1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.33 (SLES 10),
5 2.6.18-92.1.22.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
17 Frequency : rare, connect and disconnect target at same time
19 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
20 Details : don't call obd_disconnect under lov_lock. this long time
21 operation and can block ptlrpcd which answer to connect request.
26 Descriptoin: don't lose wakeup for imp_recovery_waitq
27 Details : recover_import_no_retry or invalidate_import and import_close can
28 both sleep on imp_recovery_waitq, but we was send only one wakeup
32 Frequency : always with long access acl
34 Descriptoin: mds can't pack reply with long acl.
35 Details : mds don't control size of acl but they limited by reint/getattr
38 Severity : enhancement
40 Description: Update to SLES10 kernel-2.6.16.60-0.33.
42 Severity : enhancement
44 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
47 Frequency : start MDS on uncleanly shutdowned MDS device
49 Descriptoin: aborting recovery hang on MDS
50 Details : don't throttle destroy RPCs for the MDT.
53 Frequency : on remount
55 Description: external journal device not working after the remount
56 Details : clear dev_rdonly flag for external journal devices in
62 Description: shutdown vs evict race
63 Details : client_disconnect_export vs connect request race.
64 if client will evicted at this time - we start invalidate
65 thread without referece to import and import can be freed
69 Frequency : rare, need acl's on inode.
71 Description: client can't handle ost additional correctly
72 Details : if ost was added after client connected to mds client can have
73 hit lnet_try_match_md ... to big messages to wide striped files.
74 in this case need teach client to handle config events about add
75 lov target and update client max ea size at that event.
77 Severity : enhancement
79 Description: Changelogs
80 Details : Changelogs are a lightweight mechanism to track filesystem
81 metadata and namespace changes. The changelog is recorded
82 permanently on the MDTs, and is periodically "consumed" / purged
83 when records are no longer needed.
85 Severity : enhancement
87 Description: compact fld format with extents
88 Details : Store range of seq rather than every seq in FLD. Seq
89 controller update FLD rather than clients. In Case of CMD, mdt0
90 has FLD, all other metadata server act as non persistent proxy
91 for FLD queries and cache fld entries in fld cache.
96 Description: don't skip ost target if they assigned to file
97 Details : Drop slow OSCs if we can, but not for requested start idx.
98 This means "if OSC is slow and it is not the requested
99 start OST, then it can be skipped, otherwise skip it only
100 if it is inactive/recovering/out-of-space.
104 Description: more cleanup in mds_lov
105 Details : not send LOV EA under replay, we can't know about they size at this
106 time. Don't allow client connect to mds before any ost connected,
107 for avoid problems with LOV EA size and returning EIO to client.
109 Severity : enhancement
111 Description: Interoperability at server side (Disk interoperability)
113 Severity : enhancement
115 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
117 Severity : enhancement
119 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
121 Severity : enhancement
123 Description: New client IO stack (CLIO).
125 Severity : enhancement
127 Description: Commit on sharing. Eliminate inter-client dependencies between
128 uncommitted transactions by doing transaction commits.
129 Thereby clients may recovery independently.
132 Frequency : Create a symlink file with a very long name
134 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
135 Details : If there is no extra space in the request for early cancels,
136 ldlm_req_handles_avail() returns 0 instead of a negative value.
138 Severity : enhancement
140 Description: Add /proc entry for import status
141 Details : The mdc, osc, and mgc import directories now have
142 an import directory that contains useful import data for debugging
145 Severity : enhancement
147 Description: Re-disable certain /proc logging
148 Details : Enable and disable client's offset_stats, extents_stats and
149 extents_stats_per_process stats logging on the fly.
152 Frequency : Only on FC kernels 2.6.22+
154 Description: oops in statahead
155 Details : Do not drop reference count for the dentry from VFS when lookup,
156 VFS will do that by itself.
158 Severity : enhancement
160 Description: Generic /proc file permissions
161 Details : Set /Proc file permissions in a more generic way to enable non-
162 root users operate on some /proc files.
166 Description: Hitting mdc_commit_close() ASSERTION
167 Details : Properly handle request reference release in
168 ll_release_openhandle().
172 Description: quota recovery deadlock during mds failover
173 Details : This patch includes att18982, att18236, att18237 in bz14840.
175 1. fix osts hang when mds does failover with quotaon
176 2. prevent watchdog storm when osts threads wait for the
181 Frequency : only patchless client
182 Description: add workaround for race between add/remove dentry from hash
184 Severity : enhancement
186 Description: Allow OST glimpses to return PW locks
190 Description: LBUG when llog conf file is full
191 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
195 Description: Prevent import from entering FULL state when server in recovery
199 Description: service mount cannot take device name with ":"
200 Details : Only when device name contains ":/" will mount treat it as
206 Description: replace ptlrpcd with the statahead thread to interpret the async
207 statahead RPC callback
211 Frequency : on recovery
212 Description: I/O failures after umount during fail back
213 Details : if client reconnected to restarted server we need join to recovery
214 instead of find server handler is changed and process self
215 eviction with cancel all locks.
217 Severity : enhancement
219 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
221 Severity : enhancement
223 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
225 Severity : enhancement
227 Description: Upcall on Lustre log has been dumped
228 Details : Allow for a user mode script to be called once a Lustre log has
229 been dumped. It passes the filename of the dumped log to the
230 script, the location of the script can be specified via
231 /proc/sys/lnet/debug_log_upcall.
236 Description: avoid idr_remove called for id which is not allocated.
237 Details : Move assigment s_dev for clustered nfs to end of initialization,
238 to avoid problem with error handling.
243 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
244 Details : When connection is reused this not moved from CONN_UNUSED_HASH
245 into CONN_USED_HASH and this prodice warning when put connection
246 again in unused hash.
249 Severity : enhancement
251 Description: File striping can now be set to use an arbitrary pool of OSTs.
253 Severity : enhancement
255 Description: Export bytes_read/bytes_write count on OSC/OST.
259 Description: Early reply size mismatch, MGC loses connection
260 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
261 the connect flags are properly negotiated.
266 Description: quotas are not honored with O_DIRECT
267 Details : all writes with the flag O_DIRECT will use grants which leads to
268 this problem. Now using OBD_BRW_SYNC to guard this.
272 Description: add quota statistics
273 Details : 1. sort out quota proc entries and proc code.
274 2. add quota statistics
276 Severity : enhancement
278 Description: enable quota support for HEAD.
282 Description: Properly propagate oinfo flags from lov to osc for statfs
283 Details : restore missing copy oi_flags to lov requests.
285 Severity : enhancement
287 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
288 : l_getgroups(8), lst(8), routerstat(8)
289 Details : included man pages for llobdstat(8), llstat(8),
290 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
292 Severity : enhancement
294 Description: configure's --enable-quota should check the
295 : kernel .config for CONFIG_QUOTA
296 Details : configure is terminated if --enable-quota is passed but
297 : no quota support is in kernel
301 Description: Remove portals compatibility
302 Details : Remove portals compatibility, not interoperable with releases
307 Description: Resolve device initialization race
308 Details : Prevent proc handler from accessing devices added to the
309 obd_devs array but yet be intialized.
311 Severity : enhancement
313 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
315 Severity : enhancement
317 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
322 Description: Using wrong pointer in osc_brw_prep_request
323 Details : Access to array[-1] can produce panic if kernel compiled with
324 CONFIG_PAGE_ALLOC enabled
328 Description: Client runs out of low memory
329 Details : Consider only lowmem when counting initial number of llap pages
333 Description: *optional* service tags registration
334 Details : if the "service tags" package is installed on a Lustre node
335 When the filesystem is mounted, a local-node service tag will
336 be created. See http://inventory.sun.com/ for more information
337 about the Service Tags asset management system.
341 Description: Kernel BUG tries to release flock
342 Details : Lustre does not destroy flock lock before last reference goes
343 away. So always drop flock locks when client is evicted and
344 perform unlock regardless of successfulness of speaking to MDS.
348 Description: add refcount protection for osc callbacks, avoid panic on shutdown
352 Description: sanity test 65a fails if stripecount of -1 is set
353 Details : handle -1 striping on filesystem in ll_dirstripe_verify
358 Description: ASSERTION(CheckWriteback(page,cmd)) failed
359 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
362 Severity : enhancement
364 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
368 Description: do not process already freed flock
369 Details : flock can possibly be freed by another thread before it reaches
370 to ldlm_flock_completion_ast.
374 Description: LBUG during stress test
375 Details : Need properly lock accesses the flock deadlock detection list.
379 Description: oops in page fault handler
380 Details : kernel page fault handler can return two special 'pages' in error
381 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
385 Description: timeout with invalidate import.
386 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
387 should be handled by ptlrpcd. This produce long age waiting and
388 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
390 Severity : enhancement
392 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
396 Description: enable MGS and MDT services start separately
397 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
398 a co-located MGS without starting the MGS, which is a complement
399 to 'nosvc' mount option.
403 Frequency : after recovery
404 Description: precreate to many object's after del orphan.
405 Details : del orphan st in oscc last_id == next_id and this triger growing
406 count of precreated objects. Set flag LOW to skip increase count
407 of precreated objects.
411 Frequency : rare, on clear nid stats
412 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
413 Details : when clean nid stats sometimes try destroy live entry,
414 and this produce panic in free.
418 Description: Stack overflow during MDS log replay
419 ease stack pressure by using a thread dealing llog_process.
423 Description: wait until IO finished before start new when do lock cancel.
424 Details : VM protocol want old IO finished before start new, in this case
425 need wait until PG_writeback is cleared until check dirty flag and
426 call writepages in lock cancel callback.
428 Severity : enhancement
430 Description: using special macro for print time and cleanup in includes.
434 Description: mds_mfd_close() ASSERTION(rc == 0)
435 Details : In mds_mfd_close(), we need protect inode's writecount change
436 within its orphan write semaphore to prevent possible races.
440 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
444 Frequency : rare, on shutdown ost
445 Description: don't hit live lock with umount ost.
446 Details : shrink_dcache_parent can be in long loop with destroy dentries,
447 use shrink_dcache_sb instead.
451 Description: don't panic with use echo client
452 Details : echo client pass NULL as client nid pointer and this produce null
457 Description: fix build on ppc32
458 Details : compile code with -m64 flag produce wrong object file for ppc32.
462 Description: add message levels for liblustreapi
466 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
467 Details : Move the 'good_osts' check before the 'total_bavail' check. This
468 will result in an -EAGAIN and in the exit call path we call
469 alloc_rr() which will with increasing aggressiveness attempt to
470 aquire precreated objects on the minimum number of required OSCs.
474 Description: Use old size assignment to avoid deadlock
475 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
476 the scheduling while holding a spinlock. We do not need locking
477 for size in ll_update_inode() because size is only updated from
478 the MDS for directories or files without objects, so there is no
479 other place to do the update, and concurrent access to such inodes
480 are protected by the inode lock.
484 Description: resolve "_IOWR redefined" build error on SLES10
488 Description: dump the memory debugging after all modules are unloaded to
489 suppress false negative in conf_sanity test 39
491 Severity : enhancement
493 Description: build kernel-ib packages for OFED 1.3 in our release cycle
498 Description: fix SLES kernel versioning
499 Details : the kernel version for our SLES 10 kernel did not include a "-"
500 before the "smp" at the end. while this was not a problem in
501 general, it did mean that software trying to use the kernel
502 version to try to detect a vendor specific kernel would fail.
503 this was most evident by the OFED build scripts.
507 Description: Don't update lov_desc members until making sure they are valid
508 Details : When updating lov_desc members via proc fs, need fix their
509 validities before doing the real update.
513 Description: don't put request into delay list while invalidate in flight.
514 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
515 import in flight. this produce timeout for invalidate and sometimes
516 can cause stale data.
520 Frequency : on ppc only
521 Description: not convert ost objects for directory because it's not exist.
522 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
526 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
528 Severity : enhancement
530 Description: Update to support 2.6.22.14 vanilla kernel.
534 Frequency : rare, on recovery
535 Description: read procfs can produce deadlock in some situation
536 Details : Holding lprocfs lock which send rpc can produce block for destroy
537 obd objects and this also block reconnect with -EALREADY. This
538 isn't fix all lprocfs bugs - but make it rare.
540 Severity : enhancement
542 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
545 Frequency : frequent on X2 node
547 Description: mdc_set_open_replay_data LBUG
548 Details : Set replay data for requests that are eligible for replay.
552 Description: lustre_mgs: operation 101 on unconnected MGS
553 Details : When MGC is disconnected from MGS long enough, MGS will evict the
554 MGC, and late on MGC cannot successfully connect to MGS and a lot
555 of the error messages complaining that MGS is not connected.
559 Frequency : on network error
560 Description: panic with double free request if network error
561 Details : mdc_finish_enqueue is finish request if any network error ocuring,
562 but it's true only for synchronus enqueue, for async enqueue
563 (via ptlrpcd) this incorrect and ptlrpcd want finish request
566 Severity : enhancement
568 Description: client-side metadata stat-ahead during readdir(directory readahead)
569 Details : perform client-side metadata stat-ahead when the client detects
570 readdir and sequential stat of dir entries therein
573 Frequency : on start mds
575 Description: Implement get_info(last_id) in obdfilter.
578 Frequency : occasional
580 Description: Correctly check stale fid, not start epoch if ost not support SOM
581 Details : open with flag O_CREATE need set old fid in op_fid3 because
582 op_fid2 was overwritten with new generated fid, but mds can answer
583 with one of these two fids and both is not stale. Setattr
584 incorrectly started an epoch and assume will be called
585 done_writing, but without SOM done_writing ever being called.
588 Frequency : rare, depends on device drivers and load
590 Description: MDS or OSS nodes crash due to stack overflow
591 Details : Code changes in 1.8.0 increased the stack usage of some functions.
592 In some cases, in conjunction with device drivers that use a lot
593 of stack the MDS (or possibly OSS) service threads could overflow
594 the stack. One change which was identified to consume additional
595 stack has been reworked to avoid the extra stack usage.
598 Frequency : occasional
600 Description: Do not fail import if osc_interpret_create gets -EAGAIN
601 Details : If osc_interpret_create got -EAGAIN it immediately exits and
602 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
603 oscc_has_objects and see OSC has no objests and call
604 oscc_internal_create to resend create request.
606 Severity : enhancement
608 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
610 Severity : enhancement
612 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
615 Frequency : very rare
617 Description: Fix replay if there is an un-replied request and open
618 Details : In some cases, older replay request will revert the
619 mcd->mcd_last_xid on MDS which is used to record the client's
622 Severity : enhancement
624 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
626 Severity : enhancement
628 Description: Add rhel5 support to HEAD.
630 Serverity : enhancement
632 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
637 Description: Don't allow skipping OSTs if index has been specified.
638 Details : Don't allow skipping OSTs if index has been specified, make
639 locking in internal create lots better.
643 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
644 Details : access to bitfield in structure is always rounded to long
645 and this produce problem with not atomic change any bit.
649 Description: Lustre make rpms failed.
650 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
651 builds Lustre rpms from tarball.
654 Frequency : rare on shutdown ost
656 Description: If llog cancel was not send before clean_exports phase, this can
657 produce deadlock in llog code.
658 Details : If llog thread has last reference to obd and call class_import_put
659 this produce deadlock because llog_cleanup_commit_master wait when
660 last llog_commit_thread exited, but this never success because was
661 called from llog_commit_thread.
665 Description: allow userland application know is lost one of stripes.
666 Details : fill lvb_blocks with error code on ost and return it to
667 application if error flag found.
671 Description: NULL lov_tgts causing MDS oops
672 Details : more safe checks for NULL lov_tgts for avoid oops.
674 Severity : enhancement
676 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
680 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
681 Details : If an OST is down the MDS will hang indefinitely in
682 obd_statfs_rqset() waiting for the statfs data. While for
683 MDS QOS usage of statfs, it should not stuck in waiting.
685 Severity : enhancement
687 Description: remote_acl support
688 Details : Support ACL-based permission check for remote user.
689 Support setfacl/getfacl for remote user with the utils
690 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
691 the system "{s,g}etfacl" utils.
693 Severity : enhancement
695 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
697 Severity : enhancement
699 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
703 Description: make mgs_setparam() handle fsname containing dash
704 Details : fsname containing a dash does not work with lctl conf_param
706 Severity : enhancement
708 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
710 Severity : enhancement
712 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
716 Frequency : rare, at shutdown
717 Description: access already free / zero obd_namespace.
718 Details : if client_disconnect_export was called without force flag set,
719 and exist connect request in flight, this can produce access to
720 NULL pointer (or already free pointer) when connect_interpret
721 store ocd flags in obd_namespace.
725 Frequency : only at startup
726 Description: not alloc memory with spinlock held.
727 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
728 if any spinlock held.
730 Severity : enhancement
732 Description: make lustre randomly fail allocating memory
733 Details : Make lustre randomly failed allocating memory for testing purpose.
735 Severity : enhancement
737 Description: lost problems with lov objid file
738 Details : Fixes some scability and access to not inited memory problems
739 in work with lov objdid file.
744 Description: lfs find does not continue on file error
745 Details : Continue other files processing when a file/dir is absent.
749 Description: Inconsistent usage of lustre_pack_reply()
750 Details : Standardize the usage of lustre_pack_reply() such that it
751 always generate a CERROR on failure.
754 Frequency : occasional
756 Description: MDS hang or stay in waiting lock
757 Details : If client receive lock with CBPENDING flag ldlm need send lock
758 cancel as separate rpc, to avoid situation when cancel request
759 can't processed due all i/o threads stay in wait lock.
763 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
767 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
771 Description: Update to SLES9 kernel-2.6.5-7.287.3.
775 Description: Update to RHEL4 latest kernel.
777 Severity : enhancement
779 Description: Build SLES10 patchless client fails
780 Details : The configure was broken by run ./configure with
781 --with-linux-obj=.... argument for patchless client. When the
782 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
783 can't search header adequately. Use absolute path such as
784 -I($LINUX)/include instead.
788 Description: interrupt oig_wait produce painc on resend.
789 Details : brw_redo_request can be used for resend requests from ptlrpcd and
790 private set, and this produce situation when rq_ptlrpcd_data not
791 copyed to new allocated request and triggered LBUG on assert
792 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
793 ptlrpcd set if request is changed and can be safety changed to use
798 Description: LASSERT_{REQ,REP}SWAB macros are buggy
799 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
800 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
801 these macros and replace them with their difinitions instead.
805 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
806 Details : Update which_patch & target file for SLES10 latest kernel.
809 Description: Remove client patches from SLES 10 kernel.
810 Details : This causes SLES 10 clients to behave as patchless clients
811 even on a Lustre-patched (server) kernel.
813 Severity : enhancement
815 Description: self-adjustable client's lru lists
816 Details : use adaptive algorithm for managing client cached locks lru
817 lists according to current server load, other client's work
818 pattern, memory activities, etc. Both, server and client
819 side namespaces provide number of proc tunables for controlling
822 Severity : enhancement
824 Description: light-weight GSS support
825 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
826 and reduce performance overhead.
828 Severity : enhancement
830 Description: Linux keyring support
831 Details : Support using service of Linux keyring for Lustre GSS internal
832 context refresh/cache mechanism.
836 Description: Fix errors in lfs documentation
837 Details : Fixes man pages
841 Description: don't use GFP_* in generic Lustre code.
842 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
846 Description: obdclass is limited by single OBD_ALLOC(idarray)
847 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
851 Description: warnings with build patchless client with vanila 2.6.19 and up
852 Details : change old ctl_table style and replace ctl_table/ctl_table_header
853 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
857 Frequency : Only for SLES
858 Description: sanity_quota fail test_1
859 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
860 file, which makes TSTID[2] inunique.
864 Frequency : Only for SLES9
865 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
866 Details : Update target/ChangeLog/which_patch .
870 Frequency : Only for test_10 in sanity-quota.sh
871 Description: a bug in quota test code
872 Details : checking a test flag in a wrong place
876 Frequency : Only for test_13 in sanity-quota.sh
877 Description: a bug in quota test script
878 Details : shouldn't check overfull allocated quota
882 Description: jbd statistics
883 Details : Port older jbd statistics patch for sles10
885 Severity : enhancement
887 Description: Add printing inode info into message about error in writepage.
892 Description: reply_lock_interpret crash due to race with it and lock cancel.
893 Details : Do not replay locks that are being cancelled. Do not reference
894 locks by their address during replay, just by their handle.
898 Frequency : When flocks are used.
899 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
900 Details : Flock locks might destroy just granted lock if it could be merged
901 with another existing flock, this is done in completion handler,
902 so teach ldlm_cli_enquque_fini that this is a valid case for
908 Description: Oops in read and write path when failing to allocate lock.
909 Details : Check if lock allocation failed and return error back.
913 Description: lstripe command fails for valid OST index
914 Details : The stripe offset is compared to lov->desc.ld_tgt_count
915 instead of lov->desc.ld_active_tgt_count.
919 Description: sanity.sh failed test 103
920 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
925 Description: df doesn't work properly if diskfs blocksize != 4K
926 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
930 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
931 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
933 Severity : enhancement
935 Description: Adaptive timeouts
936 Details : RPC timeouts adapt to changing server load and network
937 conditions to reduce resend attempts and improve recovery time.
941 Description: llapi_file_create() does not allow some changes
942 Details : add llapi_file_open() that allows specifying the mode and
943 open flags, and also returns an open file handle.
947 Description: merge and cleanup kernel patches.
948 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
952 Description: Add journal checksum support.(Kernel part)
953 Details : The journal checksum feature adds two new flags i.e
954 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
955 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
956 indicates that the commit block contains the checksum for
957 the blocks described by the descriptor blocks. Now commit
958 record can be sent to disk without waiting for descriptor
959 blocks to be written to disk. This behavior is controlled
960 using JBD2_FEATURE_ASYNC_COMMIT flag.
964 Description: OSS needs mutliple precreate threads
965 Details : Add ability to start more than one create thread per OSS.
969 Description: Sanity.sh test_65k failed: setstripe should have succeeded
970 Details : execute lfs setstripe on client
974 Description: mds_obd_create error creating tmp object
975 Details : When the user sets quota on root, llog will be affected and can't
976 create files and write files.
979 Frequency : Always on ia64 patchless client, and possibly others.
981 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
982 Details : This allows the patchless client to be loaded on architectures
987 Description: RedHat Update kernel for RHEL5
988 Details : Add the kernel config file for RHEL5.
992 Description: RedHat Update kernel for RHEL5
993 Details : Modify the kernel config file more closer RHEL5.
997 Description: Build failure against Centos5 (RHEL5)
998 Details : Define PAGE_SIZE when it isn't present.
1002 Description: test_7 on sanity-quota.sh can't be used on separate machines
1003 Details : add facet to handle it
1007 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1008 Details : with NFS, the anon dentry's parent was set to itself in
1009 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1010 corresponding dentry other than use rec->ur_name.
1012 Severity : enhancement
1014 Description: lfs setstripe enhancement
1015 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1019 Description: enable data checksumming by default
1020 Details : enable checksum by default, allow --disable-checksum
1021 configure option and "-o nochecksum" mount option. Checksums
1022 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1026 Description: lustre support for RHEL5
1027 Details : Add support for RHEL5.
1031 Description: Client eviction due to failover config
1032 Details : after a connection loss, the lustre client should attempt to
1033 reconnect to the last active server first before trying the
1034 other potential connections.
1038 Description: when mds and osts use different quota unit(32bit and 64bit),
1039 quota will be released repeatly.
1040 Details : void sending multiple quota reqs to mds, which will keep the
1041 status between the reqs.
1045 Description: osts not allocated evenly to files
1046 Details : change the condition to increase offset_idx
1049 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1051 Description: Data corruption for OSTs that are formatted larger than 2TB
1053 Details : When generating the bio request for lustre file writes the
1054 sector number would overflow a temporary variable before being
1055 used for the IO. The data reads correctly from Lustre (which
1056 will overflow in a similar manner) but other file data or
1057 filesystem metadata may be corrupted in some cases.
1061 Description: Tune the kernel for good SCSI performance.
1062 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1063 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1068 Description: rewrite ext2-derived code in obdclass/uuid.c
1069 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1070 from scratch preserving functionality.
1074 Description: "lfs find -obd UUID" prints directories
1075 Details : "lfs find -obd UUID" will return all directory names instead
1076 of just file names. It is incorrect because the directories
1077 do not reside on the OSTs.
1081 Description: use i_size_read and i_size_write in 2.6 port
1082 Details : replace inode->i_size access with i_size_read/write()
1085 Frequency : when removing large files
1087 Description: scheduling issue during removal of large Lustre files
1088 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1089 It causes scheduling issues when removing large files (17TB in the
1093 Frequency : only with liblustre clients on XT3
1095 Description: evictions taking too long
1096 Details : allow llrd to evict clients directly on OSTs
1099 Frequency : only on ppc
1101 Description: /proc/fs/lustre/devices broken on ppc
1102 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1103 We should make sure the type of 'index' is loff_t and then cast
1104 explicitly as needed below. Do not assign an explicitly cast
1108 Frequency : only for rhel5
1110 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1111 Details : Modify the target file & which_kernel.
1113 Severity : enhancement
1115 Description: omit set fsid for export NFS
1116 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1117 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1121 Frequency : Always, for kernels after 2.6.16
1122 Description: Fix warning idr_remove called for id=.. which is not allocated.
1123 Details : Last kernels save old s_dev before kill super and not allow
1124 to restore from callback - restore it before call kill_anon_super.
1128 Description: buffer overruns could theoretically occur
1129 Details : llapi_semantic_traverse() modifies the "path" argument by
1130 appending values to the end of the origin string, and a
1131 overrun may occur. Adding buffer overrun check in liblustreapi.
1135 Description: Fix error on 'ls .' at the top of the Lustre mount.
1136 Details : Don't revalidate dentry if it is a root dentry.
1140 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1141 Details : Modify vm-tunables-rhel4.patch.
1145 Description: Kernel config for 2.6.18-vanilla.
1146 Details : Modify targets/2.6-vanilla.target.in.
1147 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1148 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1149 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1150 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1154 Description: improve handling recoverable errors
1155 Details : if request processig with error which can be recoverable on server
1156 request should be resend, otherwise page released from cache and
1161 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1162 Details : Modify target file & which_patch.
1163 A flaw was found in the IA32 system call emulation provided
1164 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1165 value could be stored in the %RAX register, which could trigger an
1166 out-of-bounds system call table access. An untrusted local user
1167 could exploit this flaw to run code in the kernel
1168 (ie a root privilege escalation). (CVE-2007-4573).
1172 Description: O_DIRECT bypasses client statistics.
1173 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1174 read_ahead_stats were updated. copied the stats section from
1175 osc_send_oap_rpc() into async_internal().
1179 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1183 Description: change order of libsysio includes
1184 Details : '#include sysio.h' should always come before '#include xtio.h'
1188 Description: handle "serious error: objid * already exists" more gracefully
1189 Details : If LAST_ID value on disk is smaller than the objects existing in
1190 the O/0/d* directories, it indicates disk corruption and causes an
1191 LBUG(). If the object is 0-length, then we should use the existing
1192 object. This will help to avoid a full fsck in most cases.
1194 Severity : enhancement
1196 Description: adapt the lustre_config script to support the upgrade case
1197 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1198 server targets to 1.6 in parallel.
1202 Description: To avoid grant space > avaible space when the disk is almost
1203 full. Without this patch you might see the error "grant XXXX >
1204 available" or some LBUG about grant, when the disk is almost
1206 Details : In filter_check_grant, for non_grant cache write, we should
1207 check the left space by if (*left > ungranted + bytes), instead
1208 of (*left > ungranted), because only we are sure the left space
1209 is enough for another "bytes", then the ungrant space should be
1210 increase. In client, we should update cl_avail_grant only there
1211 is OBD_MD_FLGRANT in the reply.
1215 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1216 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1217 described in CVE-2007-4573. This problem could allow untrusted
1218 local users to gain root access.
1221 Frequency : when using O_DIRECT and quotas
1223 Description: Incorrect file ownership on O_DIRECT output files
1224 Details : block usage reported by 'lfs quota' does not take into account
1225 files that have been written with O_DIRECT.
1230 Description: touch file failed when fs is not full
1231 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1232 otherwise we can get ENOSP while fs is not full.
1236 Description: parallel lock callbacks
1237 Details : Instead of sending blocking and completion callbacks as separated
1238 requests, adding them to a set and sending in parallel.
1241 Frequency : only for Cray XT3
1242 Bugzilla : 12829/13455
1243 Description: Changing primary group doesn't change the group lustre assigns to
1245 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1246 overridden with the primary group provided by the group upcall,
1247 whereas the supplied fsgid can be trusted if it is in the list of
1248 supplementary groups returned by the group upcall.
1250 Severity : enhancement
1252 Description: Allow masking D_WARNING, D_ERROR messages from console
1253 Details : Console messages can now be disabled via lnet.printk.
1257 Description: User code with malformed file open parameter crashes client node
1258 Details : Before packing join_file req, all the related reference should be
1259 checked carefully in case some malformed flags cause fake
1260 join_file req on client.
1264 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1265 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1266 mode is equal to the granted mode. However, it should also check
1267 that LDLM_FL_CP_REQD is not set.
1271 Description: Heavy nfs access might result in deadlocks
1272 Details : After ELC code landed, it is now improper to enqueue any mds
1273 locks under och_sem, because enqueue might want to decide to
1274 cancel open locks for same inode we are holding och_sem for.
1278 Description: Client eviction while running blogbench
1279 Details : A lot of unlink operations with concurrent I/O can lead to a
1280 deadlock causing evictions. To address the problem, the number of
1281 oustanding OST_DESTROY requests is now throttled to
1282 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1283 callbacks are processed in priority.
1287 Description: enable ACLs on MDS by default
1288 Details : ACLs must be enabled on MDS by default.
1291 Frequency : PPC/PPC64 only
1293 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1294 Details : fix duplicated definitions between asm-ppc64/types.h and
1295 lustre_types.h on PPC.
1298 Frequency : PPC/PPC64 only
1300 Description: asm-ppc/segment.h does not exist
1301 Details : fix compile issue on PPC.
1305 Description: better handle error messages in extents code
1308 Frequency : RHEL4 only
1310 Description: mkfs is very slow on IA64/RHEL4
1311 Details : A performance regression has been discovered in the MPT Fusion
1312 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1313 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1314 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1316 Severity : enhancement
1318 Description: SNMP support enhancement
1319 Details : Adding total number of sampled request for an MDS node in snmp
1322 Severity : enhancement
1324 Description: Optimize ldlm waiting list processing for PR extent locks
1325 Details : When processing waiting list for read extent lock and meeting read
1326 lock that is same or wider to it that is not contended, skip
1327 processing rest of the list and immediatelly return current
1328 status of conflictness, since we are guaranteed there are no
1329 conflicting locks in the rest of the list.
1333 Description: Time out and refuse to reconnect
1334 Details : When the failover node is the primary node, it is possible
1335 to have two identical connections in imp_conn_list. We must
1336 compare not conn's pointers but NIDs, otherwise we can defeat
1337 connection throttling.
1341 Description: port llog fixes from b1_6 into HEAD
1342 Details : Port llog reference couting and some llog cleanups from b1_6
1343 (bug 10800) into HEAD, for protect from panic and access to already
1344 free llog structures.
1348 Description: Detect stride IO mode in read-ahead
1349 Details : When a client does stride read, read-ahead should detect that and
1350 read-ahead pages according to the detected stride pattern.
1354 Description: data checksumming impacts single node performance
1355 Details : add support for several checksum algorithm. Currently, only CRC32
1356 and Adler-32 are supported. The checksum type can be changed on
1357 the fly via /proc/fs/lustre/osc/*/checksum_type.
1361 Description: use adler32 for page checksums
1362 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1367 Description: build for x2 fails
1368 Details : fix compile issue on Cray systems.
1372 Description: Properly match for duplicate locks
1373 Details : Due to different lock order from skiplists code, we need to
1374 traverse entire list for now
1377 Frequency : only on PPC/SLES10
1379 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1380 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1381 be defined. Add a hack in mkfs_lustre.c to work around this compile
1386 Description: LASSERT on MDS when client holding flock lock dies
1387 Details : ldlm pool logic depends on number of granted locks equal to
1388 number of released locks which is not true for flock locks, so
1389 just exclude such locks from consideration.
1393 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1394 Details : Use fsfilt_sync() for both the whole filesystem sync and
1395 individual file sync to eliminate dangerous inode locking
1396 with I_LOCK that can lead to a deadlock.
1400 Description: performance in 1.6.3
1401 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1402 sure that 1MB requests are not fragmented by the block layer.
1404 Severity : enhancement
1406 Description: organize the server-side client stats on per-nid basis
1407 Details : Change the structure of stats under obdfilter and mds to
1415 The "uuid"s file would list the uuids of _active_ exports.
1416 And the clear entry is to clear all stats and stale nids.
1418 Severity : enhancement
1420 Description: eliminate client locks in face of contention
1421 Details : file contention detection and lockless i/o implementation
1422 for contended files.
1426 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1429 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1431 Description: MDS may lose file striping (and hence file data) in some cases
1432 Details : If there are additional extended attributes stored on the MDS,
1433 in particular ACLs, SELinux, or user attributes (if user_xattr
1434 is specified for the client mount options) then there is a risk
1435 of attribute loss. Additionally, the Lustre file striping
1436 needs to be larger than default (e.g. striped over all OSTs),
1437 and an additional attribute must be stored initially in the
1438 inode and then increase in size enough to be moved to the
1439 external attribute block (e.g. ACL growing in size) for file
1444 Description: skiplist implementation simplification
1445 Details : skiplists are used to group compatible locks on granted list
1446 that was implemented as tracking first and last lock of each
1447 lock group the patch changes that to using doubly linked lists
1451 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1452 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1453 because LASSERT on both the data supplied by a client, and the
1454 data on disk is dangerous and incorrect.
1456 Severity : enhancement
1458 Description: Slow truncate/writes to huge files at high offsets.
1459 Details : Directly associate cached pages to lock that protect those pages,
1460 this allows us to quickly find what pages to write and remove
1461 once lock callback is received.
1465 Description: more ldlm soft lockups
1466 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1467 starve other threads from the resource lock for a long time in
1468 case of long waiting queue, so change the debug level from
1469 D_OTHER to the less frequently used D_INFO.
1471 Severity : enhancement
1473 Description: add -gid, -group, -uid, -user options to lfs find
1477 Description: Hung threads in invalidate_inode_pages2_range
1478 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1479 once one is completed. As a result, some RPCs are stuck in the
1480 queue and are never sent.
1484 Description: filter threads hungs on waiting journal commit
1485 Details : Cleanup filter group llog code, then only filter group llog will
1486 be only created in the MDS/OST syncing process.
1490 Description: Procfs and llog threads access destoryed import sometimes.
1491 Details : Sync the import destoryed process with procfs and llog threads by
1492 the import refcount and semaphore.
1494 Severity : enhancement
1496 Description: openlock cache of b1_6 port to HEAD
1501 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1502 Details : If insertion of an extent fails, then discard the inode
1503 preallocation and free data blocks else it can lead to duplicate
1508 Description: don't always update ctime in ext3_xattr_set_handle()
1509 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1510 In some cases the ctime should not be updated, for example for
1511 2.0->1.8 compatibility it is necessary to delete an xattr and it
1512 should not update the ctime.
1516 Bugzilla : 15713/16362
1517 Description: Assertion in iopen_connect_dentry in 1.6.3
1518 Details : looking up an inode via iopen with the wrong generation number can
1519 populate the dcache with a disconneced dentry while the inode
1520 number is in the process of being reallocated. This causes an
1521 assertion failure in iopen since the inode's dentry list contains
1522 both a connected and disconnected dentry.
1526 Description: assertion failure in ldlm_handle2lock()
1527 Details : fix a race between class_handle_unhash() and class_handle2object()
1528 introduced in lustre 1.6.5 by bug 13622.
1533 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1534 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1539 Description: oops in sd_iostats_seq_show()
1540 Details : unloading/reloading the scsi low level driver triggers a kernel
1541 bug when trying to access the sd iostat file.
1546 Description: Kernel panics during QLogic driver reload
1547 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1548 patch, causing memory corruption.
1553 Description: journal_dev option does not work in b1_6
1554 Details : pass mount option during pre-mount.
1556 Severity : enhancement
1558 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1559 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1560 extent information of a file. It can be used to map logical blocks
1561 in a file to physical blocks in the block device.
1565 Description: LDLM soft lockups - improvement
1566 Details : It is be possible to send the lock handle along with each read
1567 or write request because the client is already doing a lock match
1568 itself so there isn't any reason the OST should have to re-do that
1572 Frequency : only on Cray X2
1574 Description: X2 build failures
1575 Details : fix build failures on Cray X2.
1579 Description: xid & resent requests
1580 Details : Initialize RPC XID from clock at startup (randomly if clock is
1583 Severity : enhancement
1585 Description: Add lustre_start utility to start or stop multiple Lustre servers
1590 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1591 Details : In case of memory pressure, list_del() can be called twice on
1592 req->rq_history_list, causing a kernel oops.
1596 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1597 Details : fix stack overflow in the distributed lock manager by defering
1598 export eviction after a failed AST to the elt thread instead of
1599 handling it in the dlm interpret routine.
1603 Description: Convert some comments to new format.
1604 Details : Update documenting comments to match doxygen conventions.
1608 Description: Grammar fixes.
1609 Details : A couple of trivial spelling fixes.
1613 Description: OSD_COUNTERS-mandatory
1614 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1615 to simplify the code.
1619 Description: simplify cmm_device freeing logic.
1620 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1624 Description: Add lockdep support to dt_object_operations locking interface.
1625 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1626 parameter indicating lock ordering. Update mdd code to use new
1631 Description: Introduce failloc constants for lockless IO tests.
1632 Details : Add two new failloc constants to test lockless IO. Only one of
1633 them in implemented---another is checked in yet to be landed
1638 Description: Add lockdep support for inode mutex.
1639 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1640 in the situations where Lustre has to lock more than one inode
1645 Description: Add optional invariants checking support.
1646 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1647 configure switch. This macro is to be used for consistency and
1648 sanity checks that are too expensive to be left in `production'
1653 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1654 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1659 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1660 Details : Use spin_lock_nested() in (the only) situation where more than
1661 one ldlm_lock is locked simultaneously. Also, fix possible
1662 dead-lock in ldlm_lock_change_resource() by enforcing particular
1667 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1668 Details : Instead of specifying each ldlm_lock call-back through separate
1669 parameter, wrap them into struct ldlm_callback_suite.
1673 Description: Kill join_lru obd method and its callers.
1674 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1675 in memory---a requirement implemented through ->o_join_lru() obd
1676 method in HEAD. Remove this method and its users.
1680 Description: Add asynchronous ldlm ENQUEUE completion handler.
1681 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1682 case that stock ldlm_completion_ast() cannot handle as it waits
1683 until lock is granted. Introduce new ldlm_completion_ast_async()
1684 for this. Also comment ldlm_completion_ast().
1688 Description: ldlm_error <-> errno conversion.
1689 Details : Add functions to map (rather arbitrary) between LDLM error codes
1690 and standard errno values. CLIO needs this to prevent LDLM specific
1691 constants from escaping ldlm and osc.
1695 Description: Kill unused ldlm_handle2lock_ns() function.
1696 Details : Kill unused ldlm_handle2lock_ns() function.
1700 Description: Add lu_ref support to ldlm_lock
1701 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1702 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1703 and ldlm_resource. LDLM interface has to be changed, because code that
1704 releases a reference on a lock, has to "know" what reference this is.
1705 In the most frequent case
1707 lock = ldlm_handle2lock(handle);
1709 LDLM_LOCK_PUT(lock);
1711 no changes are required. When any other reference (received _not_ from
1712 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1713 instead of LDLM_LOCK_PUT().
1715 Arguably, changes are pervasive, and interface requires some discipline
1716 for proper use. On the other hand, it was very instrumental in finding
1717 a few leaked lock references.
1721 Description: Add ldlm_lock_addref_try().
1722 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1723 attempts to addref a lock that might be being canceled
1728 Description: Add ldlm_weigh_callback().
1729 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1730 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1731 hard-coded `number of pages' logic.
1735 Description: Add lockdep annotations to llog code.
1736 Details : Use appropriately tagged _nested() locking calls in the places
1737 where llog takes more than one ->lgh_lock lock.
1741 Description: Add loi_kms_set().
1742 Details : Wrap kms updates into a helper function.
1746 Description: Constify instances of struct lsm_operations.
1747 Details : Constify instances of struct lsm_operations.
1751 Description: lu_conf support.
1752 Details : On a server, a file system object is uniquely identified
1753 by a fid, which is sufficient to locate and load all object
1754 state (inode). On a client, on the other hand, more data are
1755 necessary instantiate an object. Change lu_object_find() and
1756 friends to take additional `lu_conf' argument describing object.
1757 Typically this includes layout information.
1761 Description: lu_context fixes.
1762 Details : Introduce new lu_context functions that are needed on the client
1763 side, where some system threads (ptlrpcd) are shared by multiple
1764 modules, and so cannot be stopped during module shutdown.
1768 Description: Add start and stop methods to lu_device_type_operations.
1769 Details : Introduce two new methods in lu_device_type_operations, that are
1770 invoked when first instance of a given type is created and last one
1771 is destroyed respectively. This is need by CLIO.
1775 Description: Add lu_ref support to struct lu_device.
1776 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1777 track leaked references.
1781 Description: Introduce lu_kmem_descr.
1782 Details : lu_kmem_descr and its companion interface allow to create
1783 and destroy a number of kmem caches at once.
1787 Description: Fix lu_object finalization race.
1788 Details : Fix a race between lu_object_find() finding an object and its
1789 concurrent finalization. This race is (most likely) not possible
1790 on the server, but might happen on the client.
1794 Description: Introduce lu_ref interface.
1795 Details : lu_ref is a debugging module allowing to track references to
1796 a given object. It is quite cpu expensive, and has to be
1797 explicitly enabled with --enable-lu_ref. See usage description
1802 Description: Factor lu_site procfs stats into a separate function.
1803 Details : Separate lu_site stats printing code into a separate function
1804 that can be reused on a client.
1808 Description: Constify instances of struct {lu,dt,md}_device_operations.
1809 Details : Constify instances of struct {lu,dt,md}_device_operations.
1813 Description: Introduce struct md_site and move meta-data specific parts of
1814 struct lu_site here.
1815 Details : Move md-specific fields out of struct lu_site into special struct
1816 md_site, so that lu_site can be used on a client.
1820 Description: Kill mdd_lov_destroy().
1821 Details : Remove unused mdd code.
1825 Description: Add st_block checking to multistat.c.
1826 Details : Add st_block checking to multistat.c.
1830 Description: Add lu_ref support to struct obd_device.
1831 Details : Add lu_ref tracking to obd_device.
1835 Description: Kill obd_set_fail_loc().
1836 Details : Remove unused code.
1840 Description: Add special type for ptlrpc_request interpret functions.
1841 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1842 there. Actual usage will be in CLIO.
1846 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1847 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1848 doesn't work with lockdep.
1852 Description: Add rwv.c test program.
1853 Details : New testing program exercising readv(2) and writev(2) (Qian).
1857 Description: Add sendfile.c test program.
1858 Details : New testing program exercising sendfile(2) (Jay).
1862 Description: Ratelimit a message that can be very frequent.
1863 Details : Ratelimit a memory allocation failure message that can
1868 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1869 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1873 Description: Liblustre build fixes.
1874 Details : Liblustre build fixes.
1878 Description: libcfs: add cfs_{need,cond}_resched() interface.
1879 Details : libcfs: add cfs_{need,cond}_resched() definition and
1880 implementations for Linux, NT, and liblustre.
1882 Severity : enhancement
1884 Description: More exported tunables for mballoc
1885 Details : Add support for tunable preallocation window and new tunables for
1886 large/small requests
1890 Description: Detect corruption of block bitmap and checking for preallocations
1891 Details : Checks validity of on-disk block bitmap. Also it does better
1892 checking of number of applied preallocations. When corruption is
1893 found, it turns filesystem readonly to prevent further corruptions.
1897 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
1898 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
1899 window to 0. If it does hit cache miss, and read-pattern is still
1900 stride-io mode, does not reset the stride window, but also does not
1901 increase the stride window length in this case.
1905 Frequency : only for big-endian servers
1906 Description: Check if system is big-endian while mounting fs with extents feature
1907 Details : Mounting a filesystem with extents feature will fail on big-endian
1908 systems since ext3-based ldiskfs is not supported on big-endian
1909 systems. This can be over-riden with "bigendian_extents" mount option.
1911 Severity : enhancement
1913 Description: The root squash functionality
1914 Details : A security feature, which is to prevent users from being able
1915 to mount lustre on their desktop, run as root, and delete
1916 all of the files in the filesystem. The goal is accomplished by
1917 remapping user id (UID) and group id (GID) of the root user to
1918 a UID and GID specified by the system administartor via Lustre
1919 configuration management server (MGS). The functionality also
1920 allows to specify sets of clients for which the remapping does
1925 Description: Excessive recovery window
1926 Details : With AT enabled, the recovery window can be excessively long (6000+
1927 seconds). To address this problem, we no longer use
1928 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1929 timeout no longer depends on the service time, it is set to
1930 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1931 time via pb_service_time.
1935 Description: Watchdog triggered on MDS failover
1936 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1937 the OSTs know that the MDS "UUID" can be reused for the same export
1938 from a different NID, so we do not need to wait for the export to be
1942 Frequency : rare, only if using MMP with Linux RAID
1944 Description: MMP doesn't work with Linux RAID
1945 Details : While using HA for Lustre servers with Linux RAID, it is possible
1946 that MMP will not detect multiple mounts. To make this work we
1947 need to unplug the device queue in RAID when the MMP block is being
1948 written. Also while reading the MMP block, we should read it from
1949 disk and not the cached one.
1951 --------------------------------------------------------------------------------
1953 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1955 * Support for kernels:
1956 2.6.5-7.283 (SLES 9),
1957 2.6.9-55.EL (RHEL 4),
1958 2.6.16.46-0.14 (SLES 10),
1959 2.6.18.8 vanilla (kernel.org)
1960 * Client support for unpatched kernels:
1961 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1962 2.6.16 - 2.6.22 vanilla (kernel.org)
1963 * Due to recently discovered recovery problems, we do not recommend
1964 using patchless RHEL 4 clients with this or any earlier release.
1965 * Recommended e2fsprogs version: 1.39.cfs8
1966 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1967 * Starting with this release, the ldiskfs backing filesystem required
1968 by Lustre is now in its own package, lustre-ldiskfs. This package
1969 should be installed. It is versioned separately from Lustre and
1970 may be released separately in future.
1974 Description: block reactivating mgc import until all deactivates complete
1975 Details : Fix race when failing back MDT/MGS to itself (testing)
1978 Frequency : at statup only
1980 Description: mds_lov_synchronize race leads to various problems
1981 Details : simultaneous MDT->OST connections at startup can cause the
1982 sync to abort, leaving the OSC in a bad state.
1984 Severity : enhancement
1986 Description: add optional extra BUILD_VERSION info
1987 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1988 to override the lustre version.
1991 Frequency : 2.6.18 servers only
1993 Description: ll_kern_mount() doesn't release the module reference
1994 Details : The ldiskfs module reference count never drops down to 0
1995 because ll_kern_mount() doesn't release the module reference.
2000 Description: server LBUG when using old ost_num_threads parameter
2001 Details : Accept the old ost_num_threads parameter but warn that it
2002 is deprecated, and fix an off-by-one error that caused an LBUG.
2007 Description: Transient SCSI error results in persistent IO issue
2008 Details : iobuf->dr_error is not reinitialized to 0 between two
2012 Frequency : sometimes when underlying device returns I/O errors
2014 Description: OSTs not going read-only during write failures
2015 Details : OSTs are not remounted read-only when the journal commit threads
2016 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2017 instead of the ext3 wrappers.
2021 Description: poor connect scaling with increasing client count
2022 Details : Don't run filter_grant_sanity_check for more than 100 exports
2023 to improve scaling for large numbers of clients.
2026 Frequency : SLES10 only
2028 Description: sanity-quota.sh quotacheck failed: rc = -22
2029 Details : Quotas cannot be enabled on SLES10.
2032 Frequency : liblustre clients only
2034 Description: getdirentries does not give error when run on compute nodes
2035 Details : getdirentries does not fail when the size specified as an argument
2036 is too small to contain at least one entry
2038 Severity : enhancement
2040 Description: Add LNET router traceability for debug purposes
2041 Details : If a checksum failure occurs with a router as part of the
2042 IO path, the NID of the last router that forwarded the bulk data
2043 is printed so it can be identified.
2048 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2049 Details : Due to a race condition, liblustre clients were occasionally
2050 evicted incorrectly.
2052 Severity : enhancement
2054 Description: lfs setstripe use optional parameters instead of postional
2057 Severity : enhancement
2059 Description: Nanosecond timestamp support for ldiskfs
2060 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2061 resolution timestamps. There is not yet support for this at
2062 the Lustre filesystem level.
2065 Frequency : during server recovery
2067 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2068 Details : request with rq_no_resend flag not awake l_wait_event if they get
2072 Frequency : nfs export on patchless client
2074 Description: connectathon hang when test nfs export over patchless client
2075 Details : Disconnected dentry cannot be found with lookup, so we do not need
2076 to unhash it or make it invalid
2079 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2080 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2081 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2086 Description: open req refcounting wrong on reconnect
2087 Details : If reconnect happened between getting open reply from server and
2088 call to mdc_set_replay_data in ll_file_open, we will schedule
2089 replay for unreferenced request that we are about to free.
2090 Subsequent close will crash in variety of ways.
2091 Check that request is still eligible for replay in
2092 mdc_set_replay_data().
2097 Description: disable writes to filesystem when reading health_check file
2098 Details : the default for reading the health_check proc file has changed
2099 to NOT do a journal transaction and write to disk, because this
2100 can cause reads of the /proc file to hang and block HA state
2101 checking on a healthy but otherwise heavily loaded system. It
2102 is possible to return to the previous behaviour during configure
2103 with --enable-health-write.
2105 Severity : enhancement
2107 Description: 64-bit inode version
2108 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2109 to the inode. This will be required for version-based recovery.
2114 Description: MDS fails to start if a duplicate client export is detected
2115 Details : in some rare cases it was possible for a client to connect to
2116 an MDS multiple times. Upon recovery the MDS would detect this
2117 and fail during startup. Handle this more gracefully.
2119 Severity : enhancement
2121 Description: Add -o localflock option to simulate old noflock
2123 Details : This will achieve local-only flock/fcntl locks
2129 Description: log_commit_thread vs filter_destroy race leads to crash
2130 Details : Take import reference before releasing llog record semaphore
2135 Description: Wrong request locking in request set processing
2136 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2137 imp_delayed_list, in this place should be used imp_lock.
2140 Frequency : when reconnection
2142 Description: Grant Leak when osc reconnect to OST
2143 Details : When osc reconnect ost, OST(filter) should check whether it
2144 should grant more space to client by comparing fed_grant and
2145 cl_avail_grant, and return the granted space to client instead
2146 of "new granted" space, because client will call osc_init_grant
2147 to update the client grant space info.
2150 Frequency : when client reconnect to OST
2152 Description: Grant Leak when osc do resend and replay bulk write
2153 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2154 bulk write request, because the grant info while be sync between
2155 OSC and OST when reconnect, and we should ignore the grant info
2156 these of resend/replay write req.
2161 Description: Grant space more than avaiable left space sometimes.
2162 Details : When then OST is about to be full, if two bulk writing from
2163 different clients came to OST. Accord the avaliable space of the
2164 OST, the first req should be permitted, and the second one
2165 should be denied by ENOSPC. But if the seconde arrived before
2166 the first one is commited. The OST might wrongly permit second
2167 writing, which will cause grant space > avaiable space.
2170 Frequency : when client is evicted
2172 Description: Grant might be wrongly erased when osc is evicted by OST
2173 Details : when the import is evicted by server, it will fork another
2174 thread ptlrpc_invalidate_import_thread to invalidate the
2175 import, where the grant will be set to 0. While the original
2176 thread will update the grant it got when connecting. So if
2177 the former happened latter, the grant will be wrongly errased
2178 because of this race.
2183 Description: Checking Stale with correct fid
2184 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2185 to check whether it is stale, because sometimes, we want the
2186 enqueue happened anyway, and op_data.fid2 will not be initialized.
2188 Severity : enhancement
2190 Description: update patchless client
2191 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2194 Frequency : only with 2.4 kernel
2196 Description: random memory corruption
2197 Details : size of struct ll_inode_info is to big for union inode.u and this
2198 can be cause of random memory corruption.
2203 Description: Memory leak in recovery
2204 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2205 It should also check obd_fail before fsfilt_start, otherwise if
2206 fsfilt_start return -EROFS,(failover mds during mds recovery).
2207 then the req will return with repmsg->transno = 0 and rc = EROFS.
2208 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2209 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2210 be freed no matter whether fsfilt_commit success or not.
2213 Frequency : only with huge count clients
2215 Description: Prevents from taking the superblock lock in llap_from_page for
2217 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2218 for llap_from_page call in ll_removepage prevents from taking the
2219 superblock lock for a soon died page.
2224 Description: Not check open intent error before release open handle
2225 Details : in some rare cases, the open intent error is not checked before
2226 release open handle, which may cause
2227 ASSERTION(open_req->rq_transno != 0), because it tries to release
2228 the failed open handle.
2233 Description: Set cat log bitmap only after create log success.
2234 Details : in some rare cases, the cat log bitmap is set too early. and it
2235 should be set only after create log success.
2239 Description: Accessing a block bevice can re-enable I/O when Lustre is
2240 tearing down a device.
2241 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2246 Description: service threads may hog cpus when there are a lot of requests
2247 Details : Insert cond_resched to give other threads a chance to use some CPU
2252 Description: the cat log was not initialized in recovery
2253 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2254 unlink log on mds will not be initialized until mds post
2255 recovery. And also in mds post recovery, the unlink log will
2256 initialization will be done asynchronausly, so there will be race
2257 between add unlink log and unlink log initialization.
2261 Description: brw_stats were being printed incorrectly
2262 Details : brw_stats were being printed as log2 but all of them were not
2263 recorded as log2. Also remove some code duplication arising from
2264 filter_tally_{read,write}.
2268 Frequency : rare, only in recovery.
2269 Description: ASSERTION(req->rq_type != LI_POISON) failed
2270 Details : imp_lock should be held while iterating over imp_sending_list for
2271 prevent destroy request after get timeout in ptlrpc_queue_wait.
2275 Description: replay-single.sh test 52 fails
2276 Details : A lock's skiplist need to be cleanup when it being unlinked
2277 from its resource list.
2281 Description: Short directio read returns full requested size rather than
2283 Details : Direct I/O operations should return actual amount of bytes
2284 transferred rather than requested size.
2286 Severity : enhancement
2288 Description: metadata RPC reduction (e.g. for rm performance)
2289 Details : decrease the amount of synchronous RPC between clients and servers
2290 by canceling conflicing lock before the operation on the client
2291 and packing thier handles into the main operation RPC to server.
2293 Severity : enhancement
2295 Description: Async OSC create to avoid the blocking unnecessarily.
2296 Details : If a OST has no remain object, system will block on the creating
2297 when need to create a new object on this OST. Now, ways use
2298 pre-created objects when available, instead of blocking on an
2299 empty osc while others are not empty. If we must block, we block
2300 for the shortest possible period of time.
2304 Description: improve handling recoverable errors
2305 Details : if request processig with error which can be recoverable on server
2306 request should be resend, otherwise page released from cache and
2309 Severity : enhancement
2311 Description: refine locking for avoid write wrong info into lov_objid file
2312 Details : fix possible races with add new target and write/update data in
2315 --------------------------------------------------------------------------------
2317 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2322 Frequency : on some architectures
2324 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2325 Details : Uninitialized flags sometimes cause configuration commands to
2329 Frequency : patchless clients only
2331 Description: missing __iget() symbol export
2332 Details : The __iget() symbol export is missing. To avoid the need for
2333 this on patchless clients the deathrow inode reaper is turned
2334 off, and we depend on the VM to clean up old inodes. This
2335 dependency was during via the fix for bug 12181.
2337 --------------------------------------------------------------------------------
2339 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2341 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2342 INTEROPERATE with older versions automatically. In many cases a
2343 special upgrade step is needed. Please read the
2344 user documentation before upgrading any part of a live system.
2345 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2346 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2347 * WARNING: Lustre configuration and startup changes are required with
2348 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2352 Severity : enhancement
2354 Description: MountConf
2355 Details : Lustre configuration is now managed via mkfs and mount
2356 commands instead of lmc and lconf. New obd types (MGS, MGC)
2357 are added for dynamic configuration management. See
2358 https://mail.clusterfs.com/wikis/lustre/MountConf for
2361 Severity : enhancement
2363 Description: dynamic OST addition
2364 Details : OSTs can now be added to a live filesystem
2366 Severity : enhancement
2368 Description: startup order invariance
2369 Details : MDTs and OSTs can be started in any order. Clients only
2370 require the MDT to complete startup.
2372 Severity : enhancement
2374 Description: parallel, asynchronous orphan cleanup
2375 Details : orphan cleanup is now performed in separate threads for each
2376 OST, allowing parallel non-blocking operation.
2378 Severity : enhancement
2380 Description: optimized stripe assignment
2381 Details : stripe assignments are now made based on ost space available,
2382 ost previous usage, and OSS previous usage, in order to try
2383 to optimize storage space and networking resources.
2385 Severity : enhancement
2387 Description: Permanently set tunables
2388 Details : All writable /proc/fs/lustre tunables can now be permanently
2389 set on a per-server basis, at mkfs time or on a live system.
2391 Severity : enhancement
2393 Description: Lustre message v2
2394 Details : Add lustre message format v2.
2396 Severity : enhancement
2398 Description: client OST exclusion list
2399 Details : Clients can be started with a list of OSTs that should be
2400 declared "inactive" for known non-responsive OSTs.
2404 Description: SPEC SFS validation failure on NFS v2 over lustre.
2405 Details : Changes the blocksize for regular files to be 2x RPC size,
2406 and not depend on stripe size.
2408 Severity : enhancement
2410 Description: Multiple MD RPCs in flight.
2411 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2412 To avoid overly-overloading MDS, introduce a limit on number of
2413 MDS RPCs in flight for a single client and add /proc controls
2414 to adjust this limit.
2416 Severity : enhancement
2418 Description: client read/write statistics
2419 Details : Add client read/write call usage stats for performance
2420 analysis of user processes.
2421 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2422 file access. extents_stats shows chunk size distribution.
2423 extents_stats_per_process show chunk size distribution per
2426 Severity : enhancement
2428 Description: mds statistics
2429 Details : Add detailed mds operations statistics in
2430 /proc/fs/lustre/mds/*/stats.
2434 Description: Failure of copying files with lustre special EAs.
2435 Details : Client side always return success for setxattr call for lustre
2436 special xattr (currently only "trusted.lov").
2441 Description: Refcount LNET uuids
2442 Details : The global LNET uuid list grew linearly with every startup;
2443 refcount repeated list entries instead of always adding to
2446 Severity : enhancement
2448 Description: Dynamic service threads
2449 Details : Within a small range, start extra service threads
2450 automatically when the request queue builds up.
2453 Frequency : mixed-endian client/server environments
2455 Description: mixed-endian crashes
2456 Details : The new msg_v2 system had some failures in mixed-endian
2459 Severity : enhancement
2461 Description: Easy OST removal
2462 Details : OSTs can be permanently deactivated with e.g. 'lctl
2463 conf_param lustre-OST0001.osc.active=0'
2465 Severity : enhancement
2467 Description: MGS proc entries
2468 Details : Added basic proc entries for the MGS showing what filesystems
2471 Severity : enhancement
2473 Description: provide MGS failover
2474 Details : Added config lock reacquisition after MGS server failover.
2476 Severity : enhancement
2478 Description: add Linux 2.4 support
2479 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2483 Description: a large application tries to do I/O to the same resource and dies
2484 in the middle of it.
2485 Details : Check the req->rq_arrival time after the call to
2486 ost_brw_lock_get(), but before we do anything about
2487 processing it & sending the BULK transfer request. This
2488 should help move old stale pending locks off the queue as
2489 quickly as obd_timeout.
2492 Frequency : when an incorrect nid is specified during startup
2494 Description: ptlrpc connect to non-existant node causes kernel crash
2495 Details : LNET can't be re-entered from an event callback, which
2496 happened when we expire a message after the export has been
2497 cleaned up. Instead, hand the zombie cleanup off to another
2500 Severity : enhancement
2502 Description: plain/inodebits lock performance improvement
2503 Details : Grouping plain/inodebits in granted list by their request modes
2504 and bits policy, thus improving the performance of search through
2508 Frequency : only if OST filesystem is corrupted
2510 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2511 Details : for a short time RPCs with bulk IO are in the replay list,
2512 but replay of bulk IOs is unimplemented. If the OST filesystem
2513 is corrupted due to disk cache incoherency and then replay is
2514 started it is possible to trip an assertion. Avoid putting
2515 committed RPCs into the replay list at all to avoid this issue.
2518 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2520 Description: System hang on startup
2521 Details : This bug allowed the liblustre (e.g. catamount) client to
2522 return to the app before handling all startup RPCs. This
2523 could leave the node unresponsive to lustre network traffic
2524 and manifested as a server ptllnd timeout.
2526 Severity : enhancement
2528 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2529 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2530 boolean to control whether to print peer debug info when a
2531 client's RPC times out.
2534 Frequency : only for kernels with patches from Lustre below 1.4.3
2536 Description: Remove old rdonly API
2537 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2540 Frequency : only for devices with external journals
2542 Description: Set external device read-only also
2543 Details : During a commanded failover stop, we set the disk device
2544 read-only while the server shuts down. We now also set any
2545 external journal device read-only at the same time.
2548 Frequency : when upgrading from 1.4 while trying to change parameters
2550 Description: The wrong (new) MDC name was used when setting parameters for
2551 upgraded MDT's. Also allows changing of OSC (and MDC)
2552 parameters if --writeconf is specified at tunefs upgrade time.
2555 Frequency : when setting specific ost indicies
2557 Description: QOS code breaks on skipped indicies
2558 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2559 created with --index need not be sequential.
2561 Severity : enhancement
2563 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2564 Details : The uninit_groups feature works in conjunction with the kernel
2565 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2566 up the pass1 processing of e2fsck. This is a read-only feature
2567 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2568 mount filesystems that have had this feature enabled.
2570 Severity : enhancement
2572 Description: Improve multi-block allocation algorithm to avoid fragmentation
2573 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2574 allocation locality and avoid filesystem fragmentation.
2576 ------------------------------------------------------------------------------
2578 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2580 * Support for kernels:
2581 2.4.21-47.0.1.EL (RHEL 3)
2582 2.6.5-7.283 (SLES 9)
2583 2.6.9-42.0.10.EL (RHEL 4)
2584 2.6.12.6 vanilla (kernel.org)
2585 2.6.16.27-0.9 (SLES 10)
2586 * Recommended e2fsprogs version: 1.39.cfs5
2588 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2592 Frequency : occasional, depends on client load and configuration
2593 Bugzilla : 12181, 12203
2594 Description: data loss for recently-modified files
2596 Details : In some cases it is possible that recently written or created
2597 files may not be written to disk in a timely manner (this should
2598 normally be within 30s unless client IO load is very high).
2599 The problem appears as zero-length files or files that are a
2600 multiple of 1MB in size after a client crash or client eviction
2601 that are missing data at the end of the file.
2603 This problem is more likely to be hit on clients where files are
2604 repeatedly created and unlinked in the same directory, clients
2605 have a large amount of RAM, have many CPUs, the filesystem has
2606 many OSTs, the clients are rebooted frequently, and/or the files
2607 are not accessed by other nodes after being written.
2609 The presence of the problem can be detected by looking at
2610 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2611 smaller than the second (nr_unused) then dirty files will not
2612 be flushed automatically to disk. "sync; sleep 10" should be
2613 run several times on the node before unmounting it to update
2614 Lustre (this is also safe to run on nodes without this problem).
2616 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2617 that can cause this same problem, so customers using that kernel
2618 also need to update the kernel in addition to Lustre. In order
2619 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2621 It is normal that files written just before a client crash (less
2622 than 30s) may not yet have been flushed to disk, even for local
2626 Frequency : frequent on thin XT3 nodes
2628 Description: UUID collision on thin XT3 Linux nodes
2629 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2630 randomly, since we relied on an insufficiently-seeded PRNG.
2635 Description: OSS hangs after "All ost request buffers busy"
2636 Details : A deadlock between quota and journal operations caused OSS
2637 hangs after printing "All ost request buffers busy."
2640 Frequency : always on liblustre builds
2642 Description: Cleanup compiler warnings on liblustre
2645 Frequency : always on liblustre builds on XT3
2647 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2651 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2652 > 32000 subdirectories
2653 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2654 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2655 subdirectory count crosses 32000. This will aid e2fsck to
2656 correctly handle more than 32000 subdirectories.
2659 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2661 Description: System hang on startup
2662 Details : This bug allowed the liblustre (e.g. catamount) client to
2663 return to the app before handling all startup RPCs. This
2664 could leave the node unresponsive to lustre network traffic
2665 and manifested as a server ptllnd timeout.
2667 Severity : enhancement
2669 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2670 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2671 boolean to control whether to print peer debug info when a
2672 client's RPC times out.
2677 Description: make O_SYNC working on 2.6 kernels
2678 Details : 2.6 kernels use different method for mark pages for write,
2679 so need add a code to lustre for O_SYNC work.
2684 Description: Failure to close file and release space on NFS
2685 Details : Put inode details into lock acquired in ll_intent_file_open.
2686 Use mdc_intent_lock in ll_intent_open to properly
2687 detect all kind of errors unhandled by mdc_enqueue.
2692 Description: proc file read during shutdown sometimes raced obd removal,
2694 Details : Add lock to prevent obd access after proc file removal.
2697 Frequency : Only for files larger than 4GB on 32-bit clients.
2699 Description: improperly doing page alignment of locks
2700 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2701 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2706 Description: RPCs being resent when they shouldn't be
2707 Details : Some RPCs that should not be resent are being resent. This
2708 can cause inconsistencies in the RPC state machine. Do not
2709 resend such requests.
2712 Frequency : rare, only with NFS export
2714 Description: Crash on NFS re-export node
2715 Details : under very unusual load conditions an assertion is hit in
2716 ll_intent_file_open()
2719 Frequency : only if OST filesystem is corrupted
2721 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2722 Details : for a short time RPCs with bulk IO are in the replay list,
2723 but replay of bulk IOs is unimplemented. If the OST filesystem
2724 is corrupted due to disk cache incoherency and then replay is
2725 started it is possible to trip an assertion. Avoid putting
2726 committed RPCs into the replay list at all to avoid this issue.
2731 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2732 Details : Large single O_DIRECT read and write calls can fail to allocate
2733 a sufficiently large buffer to process the request. In case of
2734 allocation failure the allocation is retried with a smaller
2735 buffer and broken into smaller requests.
2737 Severity : enhancement
2739 Description: Add -o localflock option to simulate old noflock behaviour.
2740 Details : This will achieve local-only flock/fcntl locks coherentness.
2745 Description: versioning check is incomplete
2746 Details : Checking the version difference of client vs. server, report
2747 error if the gap is too big.
2752 Description: add support PG_writeback bit
2753 Details : add support for PG_writeback bit for Lustre, for more carefull
2754 work with page cache in 2.6 kernel. This also fix some deadlocks
2755 and remove hack for work O_SYNC with 2.6 kernel.
2757 Severity : enhancement
2759 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2760 Details : The uninit_groups feature works in conjunction with the kernel
2761 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2762 up the pass1 processing of e2fsck. This is a read-only feature
2763 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2764 mount filesystems that have had this feature enabled.
2766 Severity : enhancement
2768 Description: Improve multi-block allocation algorithm to avoid fragmentation
2769 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2770 allocation locality and avoid filesystem fragmentation.
2772 ------------------------------------------------------------------------------
2774 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2776 * Support for kernels:
2777 2.6.9-42.0.3.EL (RHEL 4)
2778 2.6.5-7.276 (SLES 9)
2779 2.4.21-47.0.1.EL (RHEL 3)
2780 2.6.12.6 vanilla (kernel.org)
2781 2.6.16.21-0.8 (SLES10)
2782 * Recommended e2fsprogs version: 1.39.cfs2-0
2784 * The backwards-compatible /proc/sys/portals symlink has been removed
2785 in this release. Before upgrading, please ensure that you change
2786 any configuration scripts or /etc/sysctl.conf files that access
2787 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2788 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2789 in advance of the upgrade on any system running Lustre 1.4.6 or
2790 newer, since /proc/sys/lnet was added in that version.
2791 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2795 Frequency : only when quota is used
2797 Description: avoid scanning export list for quota master
2798 Details : Change the algorithms to avoid scanning export list in order
2799 to improve the efficiency.
2802 Frequency : MDS failover only, very rarely
2804 Description: "went back in time" messages on mds failover
2805 Details : The greatest transno may be lost when the current operation
2806 finishes with an error (transno==0) and the client's last_rcvd
2807 record is over-written. Save the greatest transno in the
2808 mds_last_transno for this case.
2811 Frequency : always for specific kernels and striping counts
2813 Description: client may get "Matching packet too big" without ACL support
2814 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2815 when trying to access files in certain configurations. The
2816 clients should in fact be denied when mounting because they do
2817 not understand ACLs.
2820 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2822 Description: many clients connecting with IO in progress causes connect timeouts
2823 Details : Avoid synchronous journal commits to avoid delays caused by many
2824 clients connecting/disconnecting when bulk IO is in progress.
2825 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2826 OST_IO_PORTAL to avoid delays behind potentially many pending
2830 Frequency : occasionally with multiple writers to a single file
2832 Description: shared writes to file may result in wrong size reported by stat()
2833 Details : Allow growing of kms when extent lock is cancelled
2836 Frequency : always with random mmap IO to multi-striped file
2838 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2839 Details : Only if the hole is at the end of OST object so that kms is too
2840 small. Fix is to increase kms accordingly in ll_nopage.
2843 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2845 Description: writes to a missing object would leak memory on the OST
2846 Details : If there is an inconsistency between the MDS and OST filesystems,
2847 such that the MDS references an object that doesn't exist, writes
2848 to that object will leak memory due to incorrect cleanup in the
2849 error handling path, eventually running out of memory on the OST.
2854 Description: Creating too long symlink causes lustre errors
2855 Details : Check symlink and name lengths before sending requests to MDS.
2858 Frequency : only if flock is enabled (not on by default)
2860 Description: posix locks not released on fd closure on 2.6.9+
2861 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2862 kernels, this caused such locks not to be released on fd close and
2863 then assertions on fs unmount about still used locks.
2866 Frequency : MDS failover only, very rarely
2868 Description: clients may get ASSERTION(granted_lock != NULL)
2869 Details : When request was taking a long time, and a client was resending
2870 a getattr by name lock request. The were multiple lock requests
2871 with the same client lock handle and
2872 mds_getattr_name->fixup_handle_for_resent_request found one of the
2873 lock handles but later failed with ASSERTION(granted_lock != NULL).
2878 Description: handle->h_buffer_credits > 0, assertion failure
2879 Details : h_buffer_credits is zero after truncate, causing assertion
2880 failure. This patch extends the transaction or creates a new
2884 Frequency : NFS re-export or patchless client
2885 Bugzilla : 11179, 10796
2886 Description: Crash on NFS re-export node (__d_move)
2887 Details : We do not want to hash the dentry if we don't have a lock.
2888 But if this dentry is later used in d_move, we'd hit uninitialised
2889 list head d_hash, so we just do this to init d_hash field but
2890 leave dentry unhashed.
2893 Frequency : NFS re-export or patchless client
2895 Description: NFS exports has problem with symbolic link
2896 Details : lustre client didn't properly install dentry when re-exported
2897 to NFS or running patchless client.
2900 Frequency : NFS re-export or patchless client
2902 Description: Various nfs/patchless fixes.
2903 Details : fixes reuse disconected alias for lookup process - this fixes
2904 warning "find_exported_dentry: npd != pd",
2905 fix permission error with open files at nfs.
2906 fix apply umask when do revalidate.
2909 Frequency : occasional
2911 Description: Crash on NFS re-export node
2912 Details : calling clear_page() on the wrong pointer triggered oops in
2913 generic_mapping_read().
2916 Frequency : rarely, using O_DIRECT IO
2918 Description: unaligned directio crashes client with LASSERT
2919 Details : check for unaligned buffers before trying any requests.
2922 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2924 Description: stale data returned from RAID cache
2925 Details : If only a small amount of IO is done to the RAID device before
2926 reading it again it is possible to get stale data from the RAID
2927 cache instead of reading it from disk.
2930 Frequency : always for sles10 kernel
2932 Description: sles10 support
2933 Details : ll_follow_link: compile fixes and using of nd_set_link
2934 under newer kernels.
2937 Frequency : depends on arch, kernel and compiler version, always on sles10
2940 Description: recursive or deep enough symlinks cause stack overflow
2941 Details : getting rid of large stack-allocated variable in
2945 Frequency : depends on hardware
2947 Description: lustre write performance loss in the SLES10 kernel
2948 Details : the performance loss is caused by using of write barriers in the
2949 ext3 code. The SLES10 kernel turns barrier support on by
2950 default. The fix is to undo that change for ldiskfs.
2952 ------------------------------------------------------------------------------
2954 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2956 * Support for kernels:
2957 2.6.9-42.0.3EL (RHEL 4)
2958 2.6.5-7.276 (SLES 9)
2959 2.4.21-47.0.1.EL (RHEL 3)
2960 2.6.12.6 vanilla (kernel.org)
2964 Frequency : quota enabled and large files being deleted
2966 Description: releasing more than 4GB of quota at once hangs OST
2967 Details : If a user deletes more than 4GB of files on a single OST it
2968 will cause the OST to spin in an infinite loop. Release
2969 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2974 Description: statfs data retrieved from /proc may be stale or zero
2975 Details : When reading per-device statfs data from /proc, in the
2976 {kbytes,files}_{total,free,avail} files, it may appear
2977 as zero or be out of date.
2980 Frequency : systems with MD RAID1 external journal devices
2982 Description: lconf's call to blkid is confused by RAID1 journal devices
2983 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2984 of returning all block devices that match the journal UUID.
2987 Frequency : always, for aggregate stripe size over 4GB
2989 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2990 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2991 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2994 Frequency : always if "lfs find" used on a local file/directory
2996 Description: "lfs find" segfaults if used on a local file/directory
2997 Details : The case where a directory component was not specified wasn't
2998 handled correctly. Handle this properly.
3001 Frequency : always on ppc64
3003 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3004 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3005 perform endianness conversion.
3008 Frequency : rarely (truncate to non-zero file size after write under load)
3009 Bugzilla : 10730, 10687
3010 Description: Files padded with zeros to next 4K multiple
3011 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3012 it is possible that files that are truncated to a non-zero size
3013 immediately after being written are filled with zero bytes beyond
3014 the truncated size. No file data is lost.
3016 Severity : enhancement
3018 Description: Allow recovery/failover for liblustre clients.
3019 Details : liblustre clients were unaware of failover configurations until
3022 Severity : enhancement
3024 Description: user file locks should fail when not mounting with flock option
3025 Details : Set up an error-returning stub in ll_file_operations.lock field
3026 to prevent incorrect behaviour when client is mounted without
3027 flock option. Also, set up properly f_op->flock field for
3031 Frequency : always on ia64
3033 Description: "lfs df" loops on printing out MDS statfs information
3034 Details : The obd_ioctl_data was not initialized and in some systems
3035 this caused a failure during the ioctl that did not return
3036 an error. Initialize the struct and return an error on failure.
3039 Frequency : SLES 9 only
3041 Description: Error of copying files with lustre special EAs as root
3042 Details : Client side always return success for setxattr call for lustre
3043 special xattr (currently only "trusted.lov").
3046 Frequency : rarely on clusters with both ia64+i386 clients
3048 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3049 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3050 clients are doing concurrent writes to the same file it is
3051 possible that the ia64 clients may LASSERT because the OST
3052 extent locks are not PAGE_SIZE aligned. Ensure that grown
3053 locks are always aligned on the request boundary.
3056 Frequency : specific use, occasional
3058 Description: Overwriting in use executable truncates on-disk binary image
3059 Details : If one node attempts to overwrite an executable in use by
3060 another node, we now correctly return ETXTBSY instead of
3061 truncating the file.
3063 Severity : enhancement
3065 Description: Async OSC create to avoid the blocking unnecessarily.
3066 Details : If a OST has no remain object, system will block on the creating
3067 when need to create a new object on this OST. Now, ways use
3068 pre-created objects when available, instead of blocking on an
3069 empty osc while others are not empty. If we must block, we block
3070 for the shortest possible period of time.
3075 Description: chmod on Lustre root is propagated to other clients
3076 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3077 invalid by the follow_mount time.
3080 Frequency : liblustre clients only
3082 Description: Race in 'instant cancel' lock handling could lead to such locks
3083 never to be granted in case of SMP MDS
3084 Details : Do not destroy not yet granted but cbpending locks in
3088 Frequency : replay/resend of open
3090 Description: non null lock assetion failure in mds_intent_policy
3091 Details : Trying to replay/resend lockless open requests resulted in
3092 mds_open() returning 0 with no lock. Now it sets a flag if
3093 a lock is going to be returned.
3095 Severity : enhancement
3097 Description: Checksum enhancements
3098 Details : New checksum enhancements allow for resending RPCs that failed
3101 Severity : enhancement
3103 Description: Tunables on number of dirty pages in cacche
3104 Details : Allow to set limit on number of dirty pages cached.
3109 Description: client crash on unmount - lock still has references
3110 Details : In some error handling cases it was possible to leak a lock
3111 reference on a client while accessing a file. This was not
3112 harmful to the client during operation, but would cause the
3113 client to crash when the filesystem is unmounted.
3116 Frequency : specific case, rare
3118 Description: ETXTBSY on mds though file not in use by client
3119 Details : ETXTBSY is no longer incorrectly returned when attempting to
3120 chmod or chown a directory that the user previously tried to
3121 execute or a currently-executing binary.
3124 Frequency : extremely rare except on liblustre-based clients
3126 Description: Lustre space not freed when files are deleted
3127 Details : Clean up open-unlinked files after client eviction. Previously
3128 the unlink was skipped and the files remained as orphans.
3133 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3134 Details : In some cases it was possible to send a blocking callback to
3135 a client doing a glimpse, even though that client didn't get
3136 a lock granted. When the glimpse lock is cancelled on the OST
3137 the freed lock is left on the waiting list and corrupted the list.
3140 Frequency : all core dumps
3142 Description: Broke core dumps to lustre
3143 Details : Negative dentry may be unhashed if parent does not have UPDATE
3144 lock, but some callers, e.g. do_coredump, expect dentry to be
3145 hashed after successful create, hash it in ll_create_it.
3147 ------------------------------------------------------------------------------
3149 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3151 * Support for kernels:
3152 2.6.9-42.0.2.EL (RHEL 4)
3153 2.6.5-7.276 (SLES 9)
3154 2.4.21-40.EL (RHEL 3)
3155 2.6.12.6 vanilla (kernel.org)
3159 Frequency : always on RHEL 3
3161 Description: Number of open files grows over time
3162 Details : The number of open files grows over time, whether or not
3163 Lustre is started. This was due to a filp leak introduced
3164 by one of our kernel patches.
3166 ------------------------------------------------------------------------------
3168 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3170 * Support for kernels:
3171 2.6.9-42.EL (RHEL 4)
3172 2.6.5-7.276 (SLES 9)
3173 2.4.21-40.EL (RHEL 3)
3174 2.6.12.6 vanilla (kernel.org)
3179 Bugzilla : 5719, 9635, 9792, 9684
3180 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3181 Details : If a server is under heavy load and cannot reply to new
3182 connection requests before the client resends the (re)connect,
3183 the connection handling code can behave badly if two service
3184 threads are concurrently handing separate (re)connections from
3185 the same client. Add better locking to the connection handling
3186 code, and ensure that only a single connection will be processed
3187 for a given client UUID, even if the lock is dropped.
3189 Severity : enhancement
3191 Description: add TCP zero-copy support to kernel
3192 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3193 sends if the hardware supports scatter-gather and checksumming.
3194 This allows socklnd to do client-write and server-read more
3195 efficiently and reduce CPU utilization from skbuf copying.
3198 Frequency : only if NFS exporting from client
3200 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3201 Details : If setting attributes on a file created under NFS that had
3202 never been opened it would be possible to oops the client
3203 if the file had no objects.
3206 Frequency : always for liblustre
3208 Description: liblustre client does MDS+OSTs setattr RPC for each write
3209 Details : When doing a write from a liblustre client, the client
3210 incorrectly issued an RPC to the MDS and each OST the file was
3211 striped over in order to update the timestamps. When writing
3212 with small chunks and many clients this could overwhelm the MDS
3213 with RPCs. In all cases it would slow down the write because
3214 these RPCs are unnecessary.
3216 Severity : enhancement
3218 Description: allow number of MDS service threads to be changed at module load
3219 Details : It is now possible to change the number of MDS service threads
3220 running. Adding "options mds mds_num_threads={N}" to the MDS's
3221 /etc/modprobe.conf will set the number of threads for the next
3222 time Lustre is restarted (assuming the "mds" module is also
3223 reloaded at that time). The default number of threads will
3224 stay the same, 32 for most systems.
3229 Description: OST crash if filesystem is unformatted or corrupt
3230 Details : If an OST is started on a device that has never been formatted
3231 or if the filesystem is corrupt and cannot even mount then the
3232 error handling cleanup routines would dereference a NULL pointer.
3237 Description: NULL pointer deref in llap_from_page.
3238 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3239 incorrect) mapping if the page was truncated/reclaimed while it was
3240 searched for. Check for this condition and skip such pages when
3241 doing readahead. Introduce extra check to llap_from_page() to
3242 verify page->mapping->host is non-NULL (so page is not anonymous).
3245 Frequency : Sometimes when using sys_sendfile
3247 Description: "page not covered by a lock" warnings from ll_readpage
3248 Details : sendfile called ll_readpage without right page locks present.
3249 Now we introduced ll_file_sendfile that does necessary locking
3250 around call to generic_file_sendfile() much like we do in
3254 Frequency : with certain MDS communication failures at client mount time
3256 Description: NULL pointer deref after failed client mount
3257 Details : a client connection request may delayed by the network layer
3258 and not be sent until after the PTLRPC layer has timed out the
3259 request. If the client fails the mount immediately it will try
3260 to clean up before the network times out the request. Add a
3261 reference from the request import to the obd device and delay
3262 the cleanup until the network drops the request.
3265 Frequency : occasionally during client (re)connect
3267 Description: assertion failure during client (re)connect
3268 Details : processing a client connection request may be delayed by the
3269 client or server longer than the client connect timeout. This
3270 causes the client to resend the connection request. If the
3271 original connection request is replied in this interval, the
3272 client may trip an assertion failure in ptlrpc_connect_interpret()
3273 which thought it would be the only running connect process.
3276 Frequency : only with obd_echo servers and clients that are rebooted
3278 Description: kernel BUG accessing uninitialized data structure
3279 Details : When running an obd_echo server it did not start the ping_evictor
3280 thread, and when a client was evicted an uninitialized data
3281 structure was accessed. Start the ping_evictor in the RPC
3282 service startup instead of the OBD startup.
3284 Severity : enhancement
3285 Bugzilla : 10193 (patchless)
3286 Description: Remove dependency on various unexported kernel interfaces.
3287 Details : No longer need reparent_to_init, exit_mm, exit_files,
3288 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3291 Frequency : rare (only users of deprecated and unsupported LDAP config)
3293 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3294 Details : LDAP apparently contains 'lustreName' attributes instead of
3295 'name'. A simple remapping of the name is sufficient.
3298 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3300 Description: waiting_locks_callback trips kernel BUG if client is evicted
3301 Details : Running with the dump_on_timeout debug flag turned on makes
3302 it possible that the waiting_locks_callback() can try to dump
3303 the Lustre kernel debug logs from an interrupt handler. Defer
3304 this log dumping to the expired_lock_main() thread.
3306 Severity : enhancement
3308 Description: Support NFS exporting on 2.6 kernels.
3309 Details : Implement non-rawops metadata methods for NFS server to use without
3310 changing NFS server code.
3313 Frequency : very rare (synthetic metadata workload only)
3315 Description: two racing renames might cause an MDS thread to deadlock
3316 Details : Running the "racer" program may cause one MDS thread to rename
3317 a file from being the source of a rename to being the target of
3318 a rename at exactly the same time that another thread is doing
3319 so, and the second thread has already enqueued these locks after
3320 doing a lookup of the target and is trying to relock them in
3321 order. Ensure that we don't try to re-lock the same resource.
3324 Frequency : only very large systems with liblustre clients
3326 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3327 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3328 all OSTs in parallel. This allows the eviction of stale liblustre
3329 clients to proceed much faster than if they were done in series,
3330 and also offers similar improvements for other set_info RPCs.
3335 Description: excessive CPU usage during initial read phase on client
3336 Details : During the initial read phase on a client, it would agressively
3337 retry readahead on the file, consuming too much CPU and impacting
3338 performance (since 1.4.5.8). Improve the readahead algorithm
3339 to avoid this, and also improve some other common cases (read
3340 of small files in particular, where "small" is files smaller than
3341 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3346 Description: MDS crash when receiving packet with unknown intent.
3347 Details : Do not LBUG in unknown intent case, just return -EFAULT
3349 Severity : enhancement
3350 Bugzilla : 9293, 9385
3351 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3352 Details : Do not serialize getattr (non-intent version) and statfs.
3355 Frequency : occasional, when OST network is overloaded/intermittent
3357 Description: client evicted by OST after bulk IO timeout
3358 Details : If a client sends a bulk IO request (read or write) the OST
3359 may evict the client if it is unresposive to its data GET/PUT
3360 request. This is incorrect if the network is overloaded (takes
3361 too long to transfer the RPC data) or dropped the OST GET/PUT
3362 request. There is no need to evict the client at all, since
3363 the pinger and/or lock callbacks will handle this, and the
3364 client can restart the bulk request.
3367 Frequency : Always when mmapping file with no objects
3369 Description: client crashes when mmapping file with no objects
3370 Details : Check that we actually have objects in a file before doing any
3371 operations on objects in ll_vm_open, ll_vm_close and
3377 Description: Request leak when working with deleted CWD
3378 Details : Introduce advanced request refcount tracking for requests
3379 referenced from lustre intent.
3381 Severity : Enhancement
3383 Description: Cache open file handles on client.
3384 Details : MDS now will return special lock along with openhandle, if
3385 requested and client is allowed to hold openhandle, even if unused,
3386 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3387 closing files for every read/write openration.
3389 Severity : Enhancement
3391 Description: Cache open negative dentries on client when possible.
3392 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3393 negative dentries on lock revocation.
3398 Description: Remounting a client read-only wasn't possible with a zconf mount
3399 Details : It wasn't possible to remount a client read-only with llmount.
3401 Severity : enhancement
3402 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3403 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3404 MPICH is included to add Lustre-specific ADIO interfaces.
3405 This is based closely on the UFS ADIO layer and only differs
3406 in file creation, in order to allow the OST striping to be set.
3407 This is user-contributed code and not supported by CFS.
3412 Description: extended inode attributes (immutable, append-only) work improperly
3413 when 2.4 and 2.6 kernels are used on client/server or vice versa
3414 Details : Introduce kernel-independent values for these flags.
3416 Severity : enhancement
3419 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3420 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3421 granularity. Now, Fractional MB granularity is be supported,
3422 this is very useful for low memory system.
3424 Severity : enhancement
3426 Description: Getattr by fid
3427 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3428 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3429 have locks and avoids getattr rpc requests for it on every stat.
3432 Frequency : Always, for filesystems larger than 2TB
3434 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3435 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3436 filesystems larger than 2TB to be mounted with mballoc enabled.
3439 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3441 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3442 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3443 filesystems larger than 2TB is mounted then the kernel will
3444 silently corrupt the start of the filesystem. CONFIG_LBD is
3445 enabled for all CFS-supported kernels, but the possibility of
3446 this happening with a modified kernel config exists.
3448 Severity : enhancement
3450 Description: add client O_DIRECT support for 2.6 kernels
3451 Details : It is now possible to do O_DIRECT reads and writes to files
3452 in the Lustre client mountpoint on 2.6 kernel clients.
3454 Severity : enhancement
3456 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3457 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3458 parallel, not waiting for response from every OST before sending
3459 a rpc to the next OST.
3464 Description: setattr vs write race when updating file timestamps
3465 Details : Client processes that update a file timestamp into the past
3466 right after writing to the file (e.g. tar) it is possible that
3467 the updated file modification time can be reset to the current
3468 time due to a race between processing the setattr and write RPC.
3470 Severity : enhancement
3472 Description: Bring 'lfs find' closer in line with regular Linux find.
3473 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3474 -print0 options and obtains all the needed info through the lustre
3477 Severity : enhancement
3479 Description: support up to 1024 configured devices on one node
3480 Details : change obd_dev array from statically allocated to dynamically
3481 allocated structs as they are first used to reduce memory usage
3486 Description: Flush dirty partially truncated pages during truncate
3487 Details : Immediatelly flush partially truncated pages in filter_setattr,
3488 this way we completely avoid having any pages in page cache on OST
3489 and can retire ugly workarounds during writes to flush such pages.
3494 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3495 Details : protect inode from truncation within vfs_unlink() context
3496 just take a reference before calling vfs_unlink() and release it
3497 when parent's i_sem is free.
3502 Description: last_id value checked outside lock on OST caused LASSERT failure
3503 Details : If there were multiple MDS->OST object precreate requests in
3504 flight, it was possible that the OST's last object id was checked
3505 outside a lock and incorrectly tripped an assertion. Move checks
3506 inside locks, and discard old precreate requests.
3509 Frequency : always, if extents are used on OSTs
3511 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3512 Details : OSTs using the extents format would not zero the high 16 bits of
3513 the index physical block number. This is not a problem for any
3514 OST filesystems smaller than 16TB, and no kernels support ext3
3515 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3516 new/modified files) and can be fixed for existing filesystems
3517 with e2fsprogs-1.39-cfs1.
3522 Description: import connection selection may be incorrect if timer wraps
3523 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3524 import connections to be ignored if the 32-bit jiffies counter
3525 wraps. Use a 64-bit jiffies counter.
3528 Frequency : very large clusters immediately after boot
3530 Description: LNET request buffers exhausted under heavy short-term load
3531 Details : If a large number of client requests are generated on a service
3532 that has previously never seen so many requests it is possible
3533 that the request buffer growth cannot keep up with the spike in
3534 demand. Instead of dropping incoming requests, they are held in
3535 the LND until the RPC service can accept more requests.
3538 Frequency : Sometimes during replay
3540 Description: Assertion failure in ll_local_open after replay.
3541 Details : If replay happened on an open request reply before we were able
3542 to set replay handler, reply will become not swabbed tripping the
3543 assertion in ll_local_open. Now we set the handler right after
3544 recognising of open request
3547 Frequency : very rare
3549 Description: kernel reports "badness in vsnprintf"
3550 Details : Reading from the "recovery_status" /proc file in small chunks
3551 may cause a negative length in lprocfs_obd_rd_recovery_status()
3552 call to vsnprintf() (which is otherwise harmless). Exit early
3553 if there is no more space in the output buffer.
3555 Severity : enhancement
3557 Description: clear OBD RPC statistics by writing to them
3558 Details : It is now possible to clear the OBD RPC statistics by writing
3559 to the "stats" file.
3564 Description: Client mtime is not the same on different clients after utimes
3565 Details : In some cases, the client was using the utimes() syscall on
3566 a file cached on another node. The clients now validate the
3567 ctime from the MDS + OSTs to determine which one is right.
3572 Description: Inability to activate failout mode
3573 Details : lconf script incorrectly assumed that in pythong string's numeric
3574 value is used in comparisons.
3577 Frequency : always with multiple stripes per file
3579 Description: Inefficient object allocation for mutli-stripe files
3580 Details : When selecting which OSTs to stripe files over, for files with
3581 a stripe count that divides evenly into the number of OSTs,
3582 the MDS is always picking the same starting OST for each file.
3583 Return the OST selection heuristic to the original design.
3588 Description: mount failures may take full timeout to return an error
3589 Details : Under some heavy load conditions it is possible that a
3590 failed mount can wait for the full obd_timeout interval,
3591 possibly several minutes, before reporting an error.
3592 Instead return an error as soon as the status is known.
3594 Frequency : quota enabled and large files being deleted
3596 Description: releasing more than 4GB of quota at once hangs OST
3597 Details : If a user deletes more than 4GB of files on a single OST it
3598 will cause the OST to spin in an infinite loop. Release
3599 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3604 Description: statfs data retrieved from /proc may be stale or zero
3605 Details : When reading per-device statfs data from /proc, in the
3606 {kbytes,files}_{total,free,avail} files, it may appear
3607 as zero or be out of date.
3610 Frequency : systems with MD RAID1 external journal devices
3612 Description: lconf's call to blkid is confused by RAID1 journal devices
3613 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3614 of returning all block devices that match the journal UUID.
3617 Frequency : always, for aggregate stripe size over 4GB
3619 Description: assertion fails when trying to use 4GB stripe size
3620 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3621 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3624 Frequency : always on ppc64
3626 Description: the first write on an ext3 filesystem with mballoc got stuck
3627 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3628 perform endianness conversion.
3630 ------------------------------------------------------------------------------
3632 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3634 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3635 INTEROPERATE with older versions automatically. Please read the
3636 user documentation before upgrading any part of a live system.
3637 * WARNING: Lustre networking configuration changes are required with
3638 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3641 * Support for kernels:
3642 2.6.9-22.0.2.EL (RHEL 4)
3643 2.6.5-7.244 (SLES 9)
3644 2.6.12.6 vanilla (kernel.org)
3647 Severity : enhancement
3648 Bugzilla : 7981/8208
3649 Description: Introduced Lustre Networking (LNET)
3650 Details : LNET is new networking infrastructure for Lustre, it includes
3651 a reorganized network configuration mode (see the user
3652 documentation for full details) as well as support for routing
3653 between different network fabrics. Lustre Networking Devices
3654 (LNDs) for the supported network fabrics have also been
3655 created for this new infrastructure.
3657 Severity : enhancement
3658 Description: Introduced Access control lists
3659 Details : clients can set ACLs on files and directories in order to have
3660 more fine-grained permissions than the standard Unix UGO+RWX.
3661 The MDS must be started with the "-o acl" mount option.
3663 Severity : enhancement
3664 Description: Introduced filesystem quotas
3665 Details : Administrators may now establish per-user quotas on the
3668 Severity : enhancement
3670 Description: Configuration change for the XT3
3671 The PTLLND is now used to run Lustre over Portals on the XT3
3672 The configure option(s) --with-cray-portals are no longer used.
3673 Rather --with-portals=<path-to-portals-includes> is used to
3674 enable building on the XT3. In addition to enable XT3 specific
3675 features the option --enable-cray-xt3 must be used.
3680 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3681 Details : A race between allocating a new llcd and re-getting the llcd_lock
3682 allowed another thread to grab newly-allocated llcd.
3684 Severity : enhancement
3686 Description: 2.6 OST async journal commit and locking fix to improve performance
3687 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3688 2.6 kernels are now async as they already were in 2.4 kernels so
3689 that they can commit concurrently with the network bulk transfer.
3690 For block-allocated files the filter allocation semaphore is held
3691 to avoid filesystem fragmentation during allocation. BKL lock
3692 removed for 2.6 xattr operations where it is no longer needed.
3697 Description: lconf incorrectly determined whether two IP networks could talk
3698 Details : In some more complicated routing and multiple-network
3699 configurations, lconf will avoid trying to make a network
3700 connection to a disjoint part of the IP space. It was doing the
3701 math incorrectly for one set of cases.
3706 Description: Fix for potential infinite loop processing records in an llog.
3707 Details : If an llog record is corrupted/zeroed, it is possible to loop
3708 forever in llog_process(). Validate the llog record length
3709 and skip the remainder of the block on error.
3712 Frequency : occasional (liblustre only)
3714 Description: liblustre could not open files whose last component is a symlink
3715 Details : sysio_path_walk() would incorrectly pass the open intent to
3716 intermediate path components.
3719 Frequency : rare (liblustre only with non-standard tuning)
3720 Bugzilla : 7201 (7350)
3721 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3722 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3723 after it has been released, only do this for referenced locks
3725 Severity : enhancement
3727 Description: specifying an (invalid) directory default stripe_size of -1
3728 would reset the directory default striping
3729 Details : stripe_size -1 was used internally to signal directory stripe
3730 removal, now use "all default" to signal dir stripe removal
3731 as a directory striping of "all default" is not useful
3734 Frequency : common for large clusters running liblustre clients
3736 Description: doing an ls when liblustre clients are running is slow
3737 Details : sending a glimpse AST to a liblustre client waits for every AST
3738 to time out, as liblustre clients will not respond. Since they
3739 cannot cache data we refresh the OST lock LVB from disk instead.
3741 Severity : enhancement
3743 Description: doing an ls at the same time as file IO can be slow
3744 Details : enqueue and other "small" requests can be blocked behind many
3745 large IO requests. Create a new OST IO portal for non-IO
3746 requests so they can be processed faster.
3749 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3751 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3752 created file with mode 000
3753 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3754 allow an ftruncate() as a non-root user to a file with mode 000.
3755 Lustre now respects this flag to disable mode checks when
3756 truncating a file owned by the user
3759 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3762 Description: Revoking locks from clients that went dead or catatonic might take
3764 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3765 cancellation of such locks instant on servers without waiting for
3766 any reply from clients. Clients drops these locks when cancel
3767 notification from server is received without replying.
3770 Frequency : liblustre-only, when liblustre client dies or becomes busy
3772 Description: Doing ls on Linux clients can take a long time with active
3774 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3775 granting such locks to it in the first place if possible. Locks
3776 are taken by proxy on the OST during the read or write and
3777 dropped immediately afterward. Add connect flags handling, do
3778 not grant locks to liblustre clients for glimpse ASTs.
3780 Severity : enhancement
3782 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3783 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3784 sequential reads and behaves badly with applications doing random
3785 reads. Improve it by reading ahead at least read region, and
3786 avoiding excessive large RPC for small reads.
3788 Severity : enhancement
3790 Description: Creating more than 1000 files for a single job may cause a load
3791 imbalance on the OSTs if there are also a large number of OSTs.
3792 Details : qos_prep_create() uses an OST index reseed value that is an
3793 even multiple of the number of available OSTs so that if the
3794 reseed happens in the middle of the object allocation it will
3795 still utilize the OSTs as uniformly as possible.
3800 Description: OST or MDS may oops in ping_evictor_main()
3801 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3802 but doesn't restart at beginning of obd_exports_timed list
3805 Severity : enhancement
3807 Description: improve by-nid export eviction on the MDS and OST
3808 Details : allow multiple exports with the same NID to be evicted at one
3809 time without re-searching the exports list.
3812 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3814 Description: MDS may oops in groups_free()
3815 Details : in rare race conditions a newly allocated group_info struct is
3816 freed again, and this can be NULL. The 2.4 compatibility code
3817 for groups_free() checked for a NULL pointer, but 2.6 did not.
3820 Frequency : common for liblustre clients doing little filesystem IO
3821 Bugzilla : 9352, 7313
3822 Description: server may evict liblustre clients accessing contended locks
3823 Details : if a client is granted a lock or receives a completion AST
3824 with a blocking AST already set it would not reply to the AST
3825 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3828 Frequency : lfs setstripe, only systems with more than 160 OSTs
3830 Description: unable to set striping with a starting offset beyond OST 160
3831 Details : llapi_create_file() incorrectly limited the starting stripe
3832 index to the maximum single-file stripe count.
3835 Frequency : LDAP users only
3837 Description: lconf did not handle in-kernel recovery with LDAP properly
3838 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3840 Severity : enhancement
3842 Description: bind OST threads to NUMA nodes to improve performance
3843 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3844 node and do their allocations there to localize memory access
3846 Severity : enhancement
3848 Description: llmount can determine client NID directly from Myrinet (GM)
3849 Details : the client NID code from gmnalnid was moved directly into
3850 llmount, removing the need to use this or specifying the
3851 client NID explicitly when mounting GM clients with zeroconf
3854 Frequency : if client is started with down MDS
3856 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3857 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3858 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3859 ptlrpc_import_delayed_req() because we didn't check if the
3860 request was interrupted, and we also didn't break out of the
3861 event loop if there was no timeout
3866 Description: data loss during non-page-aligned writes to a single file from
3867 both multiple nodes and multiple threads on one node at same time
3868 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3869 inconsistency led to false short-reads, that were cached and later
3870 used by ->prepare_write() to fill in partially written page,
3871 leading to data loss.
3874 Frequency : always, if lconf --abort_recovery used
3876 Description: lconf --abort_recovery fails with 'Operation not supported'
3877 Details : lconf was attempting to abort recovery on the MDT device and not
3880 Severity : enhancement
3882 Description: remove cleanup logs
3883 Details : replace lconf-generated cleanup logs with lustre internal
3884 cleanup routines. Eliminates the need for client-cleanup and
3887 Severity : enhancement
3889 Description: add support for EAs (user and system) on lustre filesystems
3890 Details : it is now possible to store extended attributes in the Lustre
3891 client filesystem, and with the user_xattr mount option it
3892 is possible to allow users to store EAs on their files also
3894 Severity : enhancement
3896 Description: Add possibility (config option) to show minimal available OST free
3898 Details : When compiled with --enable-mindf configure option, statfs(2)
3899 (and so, df) will return least minimal free space available from
3900 all OSTs as amount of free space on FS, instead of summary of
3901 free spaces of all OSTs.
3903 Severity : enhancement
3905 Description: do not expand extent locks acquired on OST-side
3906 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3907 by server: they are not cached anyway.
3910 Frequency : when mmap is used/binaries executed from Lustre
3912 Description: Unmmap pages before throwing them away from read cache.
3913 Details : llap_shrink cache now attempts to unmap pages before discarding
3914 them (if unmapping failed - do not discard). SLES9 kernel has
3915 extra checks that trigger if this unmapping is not done first.
3920 Description: lconf didn't resolve symlinks before checking to see whether a
3921 given mountpoint was already in use
3924 Frequency : when migrating failover services
3925 Bugzilla : 6395, 9514
3926 Description: When migrating a subset of services from a node (e.g. failback
3927 from a failover service node) the remaining services would
3928 time out and evict clients.
3929 Details : lconf --force (implied by --failover) sets the global obd_timeout
3930 to 5 seconds in order to quickly disconnect, but this caused
3931 other RPCs to time out too quickly. Do not change the global
3932 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3934 Severity : enhancement
3935 Frequency : if MDS is started with down OST
3936 Bugzilla : 9439,5706
3937 Description: Allow startup/shutdown of an MDS without depending on the
3938 availability of the OSTs.
3939 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3940 Add appropriate locking and lov-osc refcounts for safe
3941 cleaning. Add osc abort_inflight calls in case the
3942 synchronize never started.
3945 Frequency : occasional (Cray XT3 only)
3947 Description: root not authorized to access files in CRAY_PORTALS environment
3948 Details : The client process capabilities were not honoured on the MDS in
3949 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3950 been accessed by an authorized user then root was able to access
3951 the file on the local client also. The root user capabilities
3952 are now allowed on the MDS, as this environment has secure UID.
3955 Frequency : occasional
3957 Description: ldiskfs "too long searching" message happens too often
3958 Details : A debugging message (otherwise harmless) prints too often on
3959 the OST console. This has been reduced to only happen when
3960 there are fragmentation problems on the filesystem.
3965 Description: Division by zero in statfs when all OSCs are inactive
3966 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3967 lov_statfs divides by value returned by lov_get_stripecnt().
3971 Bugzilla : 9489, 3273
3972 Description: First write from each client to each OST was only 4kB in size,
3973 to initialize client writeback cache, which caused sub-optimal
3974 RPCs and poor layout on disk for the first writen file.
3975 Details : Clients now request an initial cache grant at (re)connect time
3976 and so that they can start streaming writes to the cache right
3977 away and always do full-sized RPCs if there is enough data.
3978 If the OST is rebooted the client also re-establishes its grant
3979 so that client cached writes will be honoured under the grant.
3984 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3985 Details : Now I/O RPCs go to different portal number and (presumably) fast
3986 lock requests (and glimses) and other RPCs get their own service
3987 threads pool that should be able to service those RPCs
3990 Severity : enhancement
3992 Description: Ability to exchange lustre version between client and servers and
3993 issue warnings at client side if client is too old. Also for
3994 liblustre clients there is ability to refuse connection of too old
3996 Details : New 'version' field is added to connect data structure that is
3997 filled with version info. That info is later checked by server and
4001 Frequency : rare, liblustre only.
4002 Bugzilla : 9296, 9581
4003 Description: Two simultaneous writes from liblustre at offset within same page
4004 might proceed at the same time overwriting eachother with stale
4006 Details : I/O lock withing llu_file_prwv was released too early, before data
4007 actually was hitting the wire. Extended lock-holding time until
4008 server acknowledges receiving data.
4011 Frequency : extremely rare. Never observed in practice.
4013 Description: avoid generating lustre_handle cookie of 0.
4014 Details : class_handle_hash() generates handle cookies by incrementing
4015 global counter, and can hit 0 occasionaly (this is unlikely, but
4016 not impossible, because initial value of cookie counter is
4017 selected randonly). Value of 0 is used as a sentinel meaning
4018 "unassigned handle" --- avoid it. Also coalesce two critical
4019 sections in this function into one.
4021 Severity : enhancement
4023 Description: allow liblustre clients to delegate truncate locking to OST
4024 Details : To avoid overhead of locking, liblustre client instructs OST to
4025 take extent lock in ost_punch() on client's behalf. New connection
4026 flag is added to handle backward compatibility.
4028 Severity : enhancement
4029 Bugzilla : 4928, 7341, 9758
4030 Description: allow number of OST service threads to be specified
4031 Details : a module parameter allows the number of OST service threads
4032 to be specified via "options ost ost_num_threads={N}" in the
4033 OSS's /etc/modules.conf or /etc/modprobe.conf.
4037 Bugzilla : 6146, 9635, 9895
4038 Description: servers crash with bad pointer in target_handle_connect()
4039 Details : In rare cases when a client is reconnecting it was possible that
4040 the connection request was the last reference for that export.
4041 We would temporarily drop the export reference and get a new
4042 one, but this may have been the last reference and the export
4043 was just destroyed. Get new reference before dropping old one.
4045 Severity : enhancement
4046 Frequency : if client is started with failover MDS
4048 Description: Allow multiple MDS hostnames in the mount command
4049 Details : Try to read the configuration from all specified MDS
4050 hostnames during a client mount in case the "primary"
4053 Severity : enhancement
4055 Description: Stop sending data to evicted clients as soon as possible.
4056 Details : Check if the client we are about to send or are sending data to
4057 was evicted already. (Check is done every second of waiting,
4058 for which l_wait_event interface was extended to allow checking
4059 of exit condition at specified intervals).
4062 Frequency : rare, normally only when NFS exporting is done from client
4064 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4066 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4069 Frequency : always, if srand() is called before liblustre initialization
4071 Description: Liblustre uses system PRNG disturbing its usage by user application
4072 Details : Introduce internal to lustre fast and high-quality PRNG for
4073 lustre usage and make liblustre and some other places in generic
4074 lustre code to use it.
4076 Severity : enhancement
4077 Bugzilla : 9477, 9557, 9870
4078 Description: Verify that the MDS configuration logs are updated when xml is
4079 Details : Check if the .xml configuration logs are newer than the config
4080 logs stored on the MDS and report an error if this is the case.
4081 Request --write-conf, or allow starting with --old_conf.
4083 Severity : enhancement
4085 Description: Handle symlinks in the path when checking if Lustre is mounted.
4086 Details : Resolve intermediate symlinks when checking if a client has
4087 mounted a filesystem to avoid duplicate client mounts.
4092 Description: lconf can hit an error exception but still return success.
4093 Details : The lconf command catches the Command error exception at the top
4094 level script context and will exit with the associated exit
4095 status, but doesn't ensure that this exit status is non-zero.
4100 Description: failure of ptlrpc thread startup can cause oops
4101 Details : Starting a ptlrpc service thread can fail if there are a large
4102 number of threads or the server memory is very fragmented.
4103 Handle this without oopsing.
4106 Frequency : always, only if liblustre and non-default acceptor port was used
4108 Description: liblustre cannot connect to servers with non-default acceptor port
4109 Details : tcpnal_set_default_params() was not called and was therefore
4110 ignoring the environment varaible TCPNAL_PORT, as well as other
4111 TCPNAL_ environment variables
4116 Description: two objects could be created on the same OST for a single file
4117 Details : If an OST is down, in some cases it was possible to create two
4118 objects on a single OST for a single file. No problems other
4119 than potential performance impact and spurious error messages.
4123 Bugzilla : 5681, 9562
4124 Description: Client may oops in ll_unhash_aliases
4125 Details : Client dcache may become inconsistent in race condition.
4126 In some cases "getcwd" can fail if the current directory is
4132 Description: Inode refcounting problems in NFS export code
4133 Details : link_raw functions used to call d_instantiate without obtaining
4134 extra inode reference first.
4138 Bugzilla : 9942, 9903
4139 Description: Referencing freed requests leading to crash, memleaks with NFS.
4140 Details : We used to require that call to ll_revalidate_it was always
4141 followed by ll_lookup_it. Also with revalidate_special() it is
4142 possible to call ll_revalidate_it() twice for the same dentry
4143 even if first occurence returned success. This fix changes semantic
4144 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4145 reference on a request referred from the intent.
4146 ll_intent_release() then releases such a request.
4149 Frequency : rare, normally benchmark loads only
4151 Description: unlinked inodes were kept in memory on the client
4152 Details : If a client is repeatedly creating and unlinking files it
4153 can accumulate a lot of stale inodes in the inode slab cache.
4154 If there is no other client load running this can cause the
4155 client node to run out of memory. Instead flush old inodes
4156 from client cache that have the same inode number as a new inode.
4159 Frequency : SLES9 2.6.5 kernel and long filenames only
4160 Bugzilla : 9969, 10379
4161 Description: utime reports stale NFS file handle
4162 Details : SLES9 uses out-of-dentry names in some cases, which confused
4163 the lustre dentry revalidation. Change it to always use the
4167 Frequency : rare, unless heavy write-truncate concurrency is continuous
4168 Bugzilla : 4180, 6984, 7171, 9963, 9331
4169 Description: OST becomes very slow and/or deadlocked during object unlink
4170 Details : filter_destroy() was holding onto the parent directory lock
4171 while truncating+unlinking objects. For very large objects this
4172 may block other threads for a long time and slow overall OST
4173 responsiveness. It may also be possible to get a lock ordering
4174 deadlock in this case, or run out of journal credits because of
4175 the combined truncate+unlink. Solution is to do object truncate
4176 first in one transaction without parent lock, and then do the
4177 final unlink in a new transaction with the parent lock. This
4178 reduces the lock hold time dramatically.
4181 Frequency : rare, 2.4 kernels only
4183 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4184 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4185 before unmounting their filesystem in order to avoid deadlock.
4186 The kernel_locked() function in 2.4 kernels only checks whether
4187 the kernel lock is held, not whether it is this process that is
4188 holding it as 2.6 kernels do.
4193 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4194 Details : The client ptlrpc code may be trying to reconnect to a down
4195 server before a previous connection attempt has timed out.
4196 Increase the reconnect interval to be longer than the connection
4197 timeout interval to avoid sending duplicate connections to
4201 Frequency : echo_client brw_test command
4203 Description: fix echo_client to work with OST preallocated code
4204 Details : OST preallocation code (5137) didn't take echo_client IO path
4205 into account: echo_client calls filter methods outside of any
4206 OST thread and, hence, there is no per-thread preallocated
4207 pages and buffers to use. Solution: hijack pga pages for IO. As
4208 a byproduct, this avoids unnecessary data copying.
4212 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4213 Description: Client can oops in mdc_commit_close() after open replay
4214 Details : It was possible for the MDS to return an open request with no
4215 transaction number in mds_finish_transno() if the client was
4216 evicted, but without actually returning an error. Clients
4217 would later try to replay that open and may trip an assertion
4218 Simplify the client close codepath, and always return an error
4219 from the MDS in case the open is not successful.
4222 Frequency : rare, 2.6 OSTs only
4224 Description: OST may deadlock under high load on fragmented files
4225 Details : If there was a heavy load and highly-fragmented OST filesystems
4226 it was possible to have all the OST threads deadlock waiting on
4227 allocation of biovecs, because the biovecs were not released
4228 until the entire RPC IO was completed. Instead, release biovecs
4229 as soon as they are complete to ensure forward IO progress.
4231 Severity : enhancement
4233 Description: Support for specifying external journal device at mount
4234 Details : If an OST or MDS device is formatted with an external journal
4235 device, this device major/minor is stored in the ext3 superblock
4236 and may not be valid for failover. Allow detecting and
4237 specifying the external journal at mount time.
4242 Description: Mounting an MDS with pending unlinked files may cause oops
4243 Details : target_finish_recovery() calls mds_postrecov() which returned
4244 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4245 considers this an error and immediately begins cleaning up the
4246 lov, just after starting the mds_lov process
4248 Severity : enhancement
4250 Description: Implement 'lfs df' to report actual free space on per-OST basis
4251 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4252 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4253 usage of objects; '-h' to report in human readable format.
4255 ------------------------------------------------------------------------------
4257 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4264 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4265 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4266 at startup during orphan unlinking. Instead, skip buddy update
4267 before setup, as it will be regenerated after recovery is complete.
4272 Description: If an OST is inactive, its locks might reference stale inodes.
4273 Details : lov_change_cbdata() must iterate over all namespaces, even if
4274 they are inactive to clear inode references from the lock.
4276 Severity : enhancement
4277 Frequency : occasional, if non-standard max_dirty_mb used
4279 Description: Client will block write RPCs if not enough grant
4280 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4281 then the client will block writes while waiting for another RPC
4282 to complete instead of consuming its dirty limit. With change
4283 we get improved performance when max_dirty_mb is small.
4285 Severity : enhancement
4286 Bugzilla : 3389, 6253
4287 Description: Add support for supplementary groups on the MDS.
4288 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4289 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4290 lookups for user supplementary groups into a cache.
4294 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4295 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4296 for liblustre O_EXCL is enforced by MDS.
4300 Description: readdir never returns NULL in liblustre.
4301 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4302 dentry in struct dirent.
4306 Frequency : liblustre only, depends on application IO pattern
4307 Description: liblustre clients evicted if not contacting servers
4308 Details : Don't put liblustre clients into the ping_evictor list, so
4309 they will not be evicted by the pinger ever.
4311 Severity : enhancement
4313 Description: Add ability to evict clients by NID from MDS.
4314 Details : By echoing "nid:$NID" string into
4315 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4316 $NID would be instantly evicted from this MDS and from all active
4317 OSTs connected to it.
4321 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4322 Details : lookup_it_finish() used to query file size from OSTs that was not
4327 Description: service threads change working directory to that of init
4328 Details : Starting lustre service threads may pin the working directory
4329 of the parent thread, making that filesystem busy. Threads
4330 now change to the working directory of init to avoid this.
4334 Frequency : during shutdown only
4335 Description: shutdown with a failed MDS or OST can cause unmount to hang
4336 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4341 Frequency : 2.6 only
4342 Description: chmod/chown may include an extra supplementary group
4343 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4344 supplementary group and if none is specified this is used.
4347 Bugzilla : 5479 (6816)
4349 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4350 Details : If lookup is in progress on a file that is unlinked we might try
4351 to revalidate the inode and fail in revalidate after lookup is
4352 complete and ll_file_open() enqueues the open again but
4353 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4356 Frequency : always, if lconf --abort_recovery used
4358 Description: lconf --abort_recovery fails with 'Operation not supported'
4359 Details : lconf was attempting to abort recovery on the MDT device and not
4362 ------------------------------------------------------------------------------
4364 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4369 Frequency : rare (only unsupported configurations with a node running as an
4371 Bugzilla : 6514, 5137
4372 Description: Mounting a Lustre file system on a node running as an OST could
4374 Details : OSTs now preallocates memory needed to write out data at
4375 startup, instead of when needed, to avoid having to
4376 allocate memory in possibly low memory situations.
4377 Specifically, if the file system is mounted on on OST,
4378 memory pressure could force it to try to write out data,
4379 which it needed to allocate memory to do. Due to the low
4380 memory, it would be unable to do so and the node would
4381 become unresponsive.
4383 Severity : enhancement
4385 Description: Addition of lconf --service command line option
4386 Details : lconf now accepts a '--service <arg>' option, which is
4387 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4389 Severity : enhancement
4391 Description: Failover mode is now the default for OSTs.
4392 Details : By default, OSTs will now run in failover mode. To return to
4393 the old behaviour, add '--failout' to the lmc line for OSTs.
4395 Severity : enhancement
4397 Description: Health checks are now provided for MDS and OSTs
4398 Details : Additional detailed health check information on MSD and OSTs
4399 is now provided through the procfs health_check value.
4402 Frequency : occasional, depends on IO load
4404 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4405 after numerous create/delete cycles
4406 Details : The ext3 inode allocation policy would not allocate new inodes
4407 very well on the OSTs because there are no new directories
4408 being created. Instead we look for groups with free space if
4409 the parent directories are nearly full.
4414 Description: Network or server problems during mount may cause partially
4415 mounted clients instead of returning an error.
4416 Details : The config llog parsing code may overwrite the error return
4417 code during mount error handling, returning success instead
4423 Description: MDS can fail to allocate large reply buffers
4424 Details : After long uptimes the MDS can fail to allocate large reply
4425 buffers (e.g. zconf client mount config records) due to memory
4426 fragmentation or consumption by the buffer cache. Preallocate
4427 some large reply buffers so that these replies can be sent even
4428 under memory pressure.
4432 Frequency : rare (liblustre)
4433 Description: fsx running with liblustre complained that using truncate() to
4434 extend the file doesn't work. This patch corrects that issue.
4435 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4436 ATTR_SIZE and lsm use in llu_setattr_raw.
4440 Frequency : rare, only 2.6 kernels
4441 Description: Unusual file access patterns on the MDS may result in inode
4442 data being lost in very rare circumstances.
4443 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4444 code in the 2.6 kernel caused the inode and/or EA data not to
4445 be read from disk, causing single-file corruption.
4449 Frequency : rare, only 2.6 filesystems using extents
4450 Description: Heavy concurrent write and delete load may cause data corruption.
4451 Details : It was possible under high-load situations to have an extent
4452 metadata block in the block device cache from a just-unlinked
4453 file overwrite a newly-allocated data block. We now unmap any
4454 metadata buffers that alias just-allocated data blocks.
4458 Frequency : filesystems with default stripe_count larger than 77
4459 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4460 Details : lconf specifies an inode size of 4096 bytes when the default
4461 stripe_count is larger than 77. This conflicts with the default
4462 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4463 this case to avoid pinning too much memory for large EAs.
4465 ------------------------------------------------------------------------------
4467 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4472 Frequency : rare (extremely heavy IO load with hundreds of clients)
4474 Description: Client is evicted, gets IO error writing to file
4475 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4476 caused clients to be evicted for AST timeouts. The fixes in
4477 bug 5192 mean we no longer need to have such short AST timeouts
4478 so ldlm_timeout has been increased.
4481 Frequency : occasional during --force or --failover shutdown under load
4482 Bugzilla : 5949, 4834
4483 Description: Server oops/LBUG if stopped with --force or --failover under load
4484 Details : a collection of import/export refcount and cleanup ordering
4485 issues fixed for safer force cleanup
4488 Frequency : only filesystems larger than 120 OSTs
4489 Bugzilla : 5990, 6223
4490 Description: lfs getstripe would oops on a very large filesystem
4491 Details : lov_getconfig used kfree on vmalloc'd memory
4494 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4495 Bugzilla : 6242, 6243
4496 Description: reading from files that had been truncated to a non-zero size
4497 but never opened returned no data
4498 Details : ll_file_read() reads zeros from no-object files to EOF
4503 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4504 Details : There's little harm in aborting MDS/OSS recovery and letting it
4505 try again, so I removed the LASSERT and return an error instead.
4507 Severity : enhancement
4509 Description: New debugging infrastructure for tracking down data corruption
4510 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4511 (b) cover more of the client-side code path, and (c) try to narrow
4512 down where problems occurred
4516 Bugzilla : 3819, 4364, 4397, 6313
4517 Description: Racing close and eviction MDS could cause assertion in mds_close
4518 Details : It was possible to get multiple mfd references during close and
4519 client eviction, leading to one thread referencing a freed mfd.
4521 Severity: : enhancement
4522 Bugzilla : 3262, 6359
4523 Description: Attempts to reconnect to servers are now more aggressive.
4524 Details : This builds on the enhanced upcall-less recovery that was added
4525 in 1.4.2. When trying to reconnect to servers, clients will
4526 now try each server in the failover group every 10 seconds. By
4527 default, clients would previously try one server every 25 seconds.
4532 Description: After recovery, certain operations trigger a failed
4533 assertion on a client.
4534 Details : Failing over an mds, using lconf -d --failover, while a
4535 client was doing a readdir() call would cause the client to
4536 LBUG after recovery completed and the readdir() was resent.
4538 Severity : enhancement
4540 Description: Default groups are now added by lconf
4541 Details : You can now run lconf --group <servicename> without having to
4542 manually add groups with lmc.
4545 Frequency : occasional
4547 Description: Nodes with an elan id of 0 trigger a failed assertion
4550 Frequency : always when accessing e.g. tty/console device nodes
4552 Description: tty and some other devices nodes cannot be used on lustre
4553 Details : file's private_data field is used by device data and lustre
4554 values in there got lost. New field was added to struct file to
4555 store fs-specific private data.
4558 Frequency : when exporting Lustre via NFS
4560 Description: NFSD failed occasionally when looking up a path component
4561 Details : NFSD is looking up ".." which was broken in ext3 directories
4562 that had grown large enough to become hashed.
4565 Frequency : Clusters with multiple interfaces not on the same subnet
4567 Description: Nodes will repeatedly try to reconnect to an interface which it
4568 cannot reach and report an error to the log.
4569 Details : Extra peer list entries will be created by lconf with some peers
4570 unreachable. lconf now validates the peer before adding it.
4573 Frequency : Only if a default stripe is set on the filesystem root.
4575 Description: Setting a default stripe on the filesystem root prevented the
4576 filesystem from being remounted.
4577 Details : The client was sending extra request flags in the root getattr
4578 request and did not allocate a reply buffer for the dir EA.
4581 Frequency : occasional, higher if lots of files are accessed by one client
4582 Bugzilla : 6159, 6097
4583 Description: Client trips assertion regarding lsm mismatch/magic
4584 Details : While revalidating inodes the VFS looks up inodes with ifind()
4585 and in rare cases can find an inode that is being freed.
4586 The ll_test_inode() code will free the lsm during ifind()
4587 when it finds an existing inode and then the VFS later attaches
4588 this free lsm to a new inode.
4592 Bugzilla : 6422, 7030
4593 Description: MDS deadlock between mkdir and client eviction
4594 Details : Creating a new file via mkdir or mknod (starting a transaction
4595 and getting the ns lock) can deadlock with client eviction
4596 (gets ns lock and trying to finish a synchronous transaction).
4599 Frequency : occasional
4600 Description: While starting a server, the fsfilt_ext3 module could not be
4602 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4603 kernels. Previously, lconf would still use the ext3 name
4604 when trying to load modules. Now, it will correctly use
4605 ext3 on 2.4 and ldiskfs on 2.6.
4607 Severity : enhancement
4608 Description: The default stripe count has been changed to 1
4609 Details : The interpretation of the default stripe count (0, to lfs
4610 or lmc) has been changed to mean striping across a single
4611 OST, rather than all available. For general usage we have
4612 found a stripe count of 1 or 2 works best.
4614 Severity : enhancement
4615 Description: Add support for compiling against Cray portals.
4616 Details : Conditional compiling for some areas that are different
4620 Frequency : occasional
4621 Bugzilla : 6409, 6834
4622 Description: Creating files with an explicit stripe count may lead to
4623 a failed assertion on the MDS
4624 Details : If some OSTs are full or unavailable, creating files may
4625 trigger a failed assertion on the MDS. Now, Lustre will
4626 try to use other servers or return an error to the
4630 Frequency : occasional
4632 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4633 caused lock timeout during readahead (since 1.4.2).
4634 Details : Processes doing readahead might match a lock that hasn't been
4635 granted yet if there are overlapping and conflicting lock
4636 requests. The readahead process waits on ungranted lock
4637 (original lock is CBPENDING), while OST waits for that process
4638 to cancel CBPENDING read lock and eventually evicts client.
4640 Severity : enhancement
4642 Description: Initial enabling of flock support for clients
4643 Details : Implements fcntl advisory locking and file status functions.
4644 This feature is provided as an optional mount flag (default
4645 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4646 locking are implemented yet, and those that are are not guaranteed
4647 to be completely correct in production environments.
4648 mount -t lustre -o [flock|noflock] ...
4651 Frequency : occasional
4653 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4654 assertion in the ext3 JBD (journaling) layer.
4655 Details : The b_committed_data struct is protected by the big kernel lock
4656 in 2.4 kernels, serializing journal_commit_transaction() and
4657 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4658 access to this struct. In 2.6 kernels there is finer grained
4659 locking to improve SMP performance of the JBD layer.
4663 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4668 Description: merge and cleanup kernel patches.
4671 -----------------------------------------------------------------------------
4673 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4675 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4676 versions of Lustre. Please update all servers and clients to
4677 version 1.4.2 or later at the same time. You must also run
4678 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4679 to update the configuration logs.
4681 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4682 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4683 - swab llog records of type '0' so we get proper header size/idx (5861)
4684 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4685 - fix rename of one directory over another leaking an inode (5953)
4686 - avoid SetPageDirty on 2.6 (5981)
4687 - don't re-add just-being-destroyed locks to the waiting list (5653)
4688 - when creating new directories, inherit the parent's custom
4689 striping settings if present parent (3048)
4690 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4691 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4692 - don't hold client locks on temporary worklist from l_lru (5666)
4693 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4694 - automatically evict dead clients (5921)
4695 - Update file size properly in create+truncate+fstat case (6196)
4696 - Do not unhash mountpoint dentries, do not allow removal of
4698 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4699 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4700 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4701 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4702 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4703 - Correctly handle reads of files with no objects (6243)
4704 - lctl recover will also mark a device active if deactivate used (5933)
4706 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4707 - allow --write-conf on an MDS with different nettype than client (5619)
4708 - don't write config llogs to MDS for mounts not from that MDS (5617)
4709 - lconf should create multiple TCP connections from a client (5201)
4710 - init scripts are now turned off by default; run chkconfig --on
4711 lustre and chkconfig --on lustrefs to use them
4712 - upcalls are no longer needed for clients to recover to failover
4714 - add --abort-recovery option to lconf to abort recovery on device
4716 - add support for an arbitrary number of OSTs (3026)
4717 - Quota support protocol changes.
4718 - forward compatibility changes to wire structs (6007)
4719 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4720 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4721 - improved "lustre status" script
4722 - initialize blocksize for non-regular files (6062)
4723 - added --disable-server and --disable-client configure options (5782)
4724 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4725 - Vanilla 2.4.29 support
4726 - increase maximum number of obd devices to 520 (6242)
4727 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4728 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4729 SLES 9 2.6.5 kernels
4730 - limit stripes per file to 160 (the maximum EA size) (6093)
4732 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4735 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4736 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4737 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4738 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4739 - small ext3 extents cleanups and fixes (5733)
4740 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4741 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4742 - avoid needless client->OST connect, fix handle mismatch (5317)
4743 - fix DLM error path that led to out-of-sync client, long delays (5779)
4744 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4745 - fix several locking issues related to i_size (5492,5624,5654,5672)
4746 - don't move pending lock onto export if it is already evicted (5683)
4747 - fix kernel oops when creating .foo in unlinked directory (5548)
4748 - fix deadlock in obdfilter statistics vs. object create (5811)
4749 - use time_{before,after} to avoid timer jiffies wrap (5882)
4750 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4751 - Do not leak request if server was not able to process it (5154)
4752 - If mds_open unable to find parent dir, make that negative lookup(5154)
4753 - don't create new directories with extent-mapping (5909, 5936)
4755 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4756 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4757 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4758 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4759 - support for RHEL4 kernel on i686 (5773)
4760 - provide error messages when incompatible logs are encountered (5898)
4762 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4763 * version 1.4.0.10 (1.4.1 release candidate 1)
4765 - don't keep a lock reference when lock is not granted (4238)
4766 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4767 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4768 - drop import inflight refcount on signal_completed_replay error (5255)
4769 - unlock page after async write error during send (3677)
4770 - handle missing objects in filter_preprw_read properly (5265)
4771 - no transno return for symlink open, don't save no-trasno open (3440)
4772 - don't try to complete elan receive that already failed (4012)
4773 - free RPC server reply state on error (5406)
4774 - clean up thread from ptlrpc_start_thread() on error (5160)
4775 - readahead could read extra page into cache that wasn't ejected (5388)
4776 - prevent races in class_attach/setup/cleanup/detach (5260)
4777 - don't dereference de->d_inode after l_dput of de (5458)
4778 - use "int" for stripe value returned from lock_to_stripe (5544)
4779 - mballoc allocation and error-checking fixes in 2.6 (5504)
4780 - block device patches to fix I/O request sizes in 2.6 (5482)
4781 - look up hostnames for IB nals (5602)
4782 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4783 - don't start multiple acceptors for the same port (5277)
4784 - fix incorrect LASSERT in mds_getattr_name (5635)
4785 - export a proc file for general "ping" checking (5628)
4786 - fix "lfs check" to not block when the MDS is down (5628)
4788 - service request history (4965)
4789 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4790 - create an "evict_client" /proc entry on OSTs, like the MDS has
4791 - fix mount usage message, return errors per mount(8) (5168)
4792 - change grep [] to grep "[]" in tests so they work in more UMLs
4793 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4794 - remove ancient LOV_MAGIC_V0 EA support (5047)
4795 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4796 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4797 - allow readahead allocations to fail when low on memory (5383)
4798 - mmap locking landed again, after considerable improvement (2828)
4799 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4801 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4804 - send OST transaction number in read/write reply to free req (4966)
4805 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4806 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4807 - the watchdog thread now runs as interruptible (5246)
4808 - flock/lockf fixes (but it's still disabled, pending 5135)
4809 - don't use EXT3 constants in llite code (5094)
4810 - memory shortage at startup could cause assertion (5176)
4812 - reorganization of lov code
4813 - single portals codebase
4815 - add extents/mballoc support (5025)
4816 - direct I/O reads in the obdfilter (4048)
4817 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4819 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4822 - send OST transaction number in read/write reply to free req (4966)
4823 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4824 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4825 - the watchdog thread now runs as interruptible (5246)
4826 - handle missing objects in filter_preprw_read properly (5265)
4827 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4828 - drop import inflight refcount on signal_completed_replay error (5255)
4829 - unlock page after async write error during send (3677)
4830 - return original error code on reconstructed replies (3761)
4831 - no transno return for symlink open, don't save no-trasno open (3440)
4833 - add pid to ldlm debugging output (4922)
4834 - bump the watchdog timeouts -- we can't handle 30sec yet
4835 - extra debugging for orphan dentry/inode bug (5259)
4837 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4840 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4841 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4842 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4843 - fix NULL dereference in /proc/sys/portals/routes (4827)
4844 - allow failed mdc_close() operations to be interrupted (4561)
4845 - stop precreate on OST before MDS would time out on it (4778)
4846 - don't send partial-page writes before EOF from client (4410)
4847 - discard client grant for sub-page writes on large-page clients (4520)
4848 - don't free dentries not owned by NFS code, check generation (4806)
4849 - fix lsm leak if mds_create_objects() fails (4801)
4850 - limit debug_daemon file size, always print CERROR messages (4789)
4851 - use transno after validating reply (3892)
4852 - process timed out requests if import state changes (3754)
4853 - update mtime on OST during writes, return in glimpse (4829)
4854 - add mkfsoptions to LDAP (4679)
4855 - use ->max_readahead method instead of zapping global ra (5039)
4856 - don't interrupt __l_wait_event() during strace
4858 - add software watchdogs to catch hung threads quickly (4941)
4859 - make lustrefs init script start after nfs is mounted
4860 - fix CWARN/ERROR duplication (4930)
4861 - return async write errors to application if possible (2248)
4862 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4863 - print NAL number in %x format (4645)
4864 - update barely-supported suse-2.4.21-171 series (4842)
4865 - support for sles 9 %post scripts
4866 - support for building 2.6 kernel-source packages
4867 - support for sles km_* packages
4869 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4872 - ignore -ENOENT errors in osc_destroy (3639)
4873 - notify osc create thread that OSC is being cleaned up (4600)
4874 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4875 - reconstruct ost_handle() like mds_handle() (4657)
4876 - create a new thread to do import eviction to avoid deadlock (3969)
4877 - let lconf resolve symlinked-to devices (4629)
4878 - don't unlink "objects" from directory with default EA (4554)
4879 - hold socknal file ref over connect in case target is down (4394)
4880 - allow more than 32000 subdirectories in a single directory (3244)
4881 - fix blocks count for O_DIRECT writes (3751)
4882 - OST returns ENOSPC from object create when no space left (4539)
4883 - don't send truncate RPC if file size isn't changing (4410)
4884 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4885 - bind to privileged port in socknal and tcpnal (3689)
4887 - rate limit CERROR/CWARN console message to avoid overload (4519)
4888 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4889 - basic mmap support (3918)
4890 - kernel patch series update from b1_4 (4711)
4892 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4895 - avoid crash during MDS cleanup with OST shut down (2775)
4896 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4897 - don't use bad inodes on the MDS (3744)
4898 - dynamic object preallocation to improve recovery speed (4236)
4899 - don't hold spinlock over lock dumping or change debug flags (4401)
4900 - don't zero obd_dev when it is force cleaned (3651)
4901 - print grants to console if they go negative (4431)
4902 - "lctl deactivate" will stop automatic recovery attempts (3406)
4903 - look for existing locks in ldlm_handle_enqueue() (3764)
4904 - don't resolve lock handle twice in recovery avoiding race (4401)
4905 - revalidate should check working dir is a directory (4134)
4907 - don't always mark "slow" obdfilter messages as errors (4418)
4909 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4912 - don't close LustreDB during write_conf until it is done (3860)
4913 - fix typo in lconf for_each_profile (3821)
4914 - allow dumping logs from multiple threads at one time (3820)
4915 - don't allow multiple threads in OSC recovery (3812)
4916 - fix debug_size parameters (3864)
4917 - fix mds_postrecov to initialize import for llog ctxt (3121)
4918 - replace config semaphore with spinlock (3306)
4919 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4920 - don't allow enqueue to complete on a destroyed export (3822)
4921 - down write_lock before checking llog header bitmap (3825)
4922 - recover from lock replay timeout (3764)
4923 - up llog sem before sending rpc (3652)
4924 - reduce ns lock hold times when setting kms (3267)
4925 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4926 - fix NULL deref and obd_dev leak on setup error (3312)
4927 - replace some LBUG about llog ops with error handling (3841)
4928 - don't match INVALID dentries from d_lookup and spin (3784)
4929 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4930 - fix invalid assertion in ptlrpc_set_wait (3880)
4932 - add libwrap support for the TCP acceptor (3996)
4933 - add /proc/sys/portals/routes for non-root route listing (3994)
4934 - allow setting MDS UUID in .xml (2580)
4935 - print the stack of a process that LBUGs (4228)
4937 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4940 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4941 - make sure to unset replay flag from failed open requests (3440)
4942 - if default stripe count is 0, use OST count for inode size (3636)
4943 - update parent mtime/ctime on client for create/unlink (2611)
4944 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4945 - free recovery state on server during a forced cleanup (3571)
4946 - unregister_reply for resent reqs (3063)
4947 - loop back devices mounting and status check on 2.6 (3563)
4948 - fix resource-creation race that can provoke i_size == 0 (3513)
4949 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4950 - more debugging for page-accounting assertion (3746)
4951 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4952 - avoid deadlock after precreation failure (3758)
4953 - fix race and lock order deadlock in orphan handling (3450, 3750)
4954 - add validity checks when grabbing inodes from l_ast_data (3599)
4956 - add /proc/.../recovery_status to obdfilter (3428)
4957 - lightweight CDEBUG infrastructure, debug daemon (3668)
4958 - change default OSC RPC parameters to be better on small clusters
4959 - turn off OST read cache for files smaller than 32MB
4960 - install man pages and include them in rpms (3100)
4961 - add new init script for (un)mounting lustre filesystems (2593)
4962 - run chkconfig in %post for init scripts (3701)
4963 - drop scimac NAL (unmaintained)
4965 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4968 - clean kiobufs before and after use (3485)
4969 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4970 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4971 - change init script to fail more clearly if not run as root (1528)
4972 - allow clients to reconnect during replay (1742)
4973 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4974 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4975 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4976 - fix cond_resched (3554)
4977 - only evict unfinished clients after recovery (3515)
4978 - allow bulk resend, prevent data loss (3570)
4979 - dynamic ptlrpc request buffer allocation (2102)
4980 - don't allow unlinking open directory if it isn't empty (2904)
4981 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4982 - remove extraneous obd dereference causing LASSERT failure (3334)
4983 - don't use get_cycles() when creating temp. files on the mds (3156)
4984 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4985 - handle EEXIST for set-stripe, set proper directory name (3336)
4987 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4988 - fix ksocknal_fmb_callback() error messages (2918)
4990 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4993 - don't copy lvb into (possibly NULL) reply on error (2983)
4994 - don't deref dentry after dput, don't free lvb on error (2922)
4995 - use the kms to determine writeback rpc length (2947)
4996 - increment oti_logcookies when osc is inactive (2948)
4997 - update client's i_blocks count via lvb messages (2543)
4998 - handle intent open/close of special files properly (1557)
4999 - mount MDS with errors=remount-ro, like obdfilter (2009)
5000 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5001 - don't use cancelling-locks' kms values (2947)
5002 - use highest lock extent for kms, not last one (2925)
5003 - don't dereference ERR_PTR() dentry in error handling path (3107)
5004 - fix thread race in portals_debug_dumplog() (3122)
5005 - create lprocfs device entries at setup instead of at attach (1519)
5006 - common AST error handler, don't evict client on completion race (3145)
5007 - zero nameidata in detach_mnt in 2.6 (3118)
5008 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5009 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5010 - initialize RPC timeout timer earlier for 2.6 (3219)
5011 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5012 - print nal/nid for unknown nid (3258)
5013 - additional checks for oscc recovery before doing precreate (3284)
5014 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5015 - don't crash in mdc_close for bad permissions on open (3285)
5016 - zero i_rdev for non-device files (3147)
5017 - clear page->private before handing to FS, better assertion (3119)
5018 - tune the read pipeline (3236)
5019 - fix incorrect decref of invalidated dentry (2350)
5020 - provide read-ahead stats and refine rpc in flight stats (3328)
5021 - don't hold journal transaction open across create RPC (3313)
5022 - update atime on MDS at close time (3265)
5023 - close LDAP connection when recovering to avoid server load (3315)
5024 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5025 - don't leak open file on MDS after open resend (3325)
5026 - serialize filter_precreate and filter_destroy_precreated (3329)
5027 - loop device shouldn't call sync_dev() for nul device (3092)
5028 - clear page cache after eviction (2766)
5029 - resynchronize MDS->OST in background (2824)
5030 - refuse to mount the same filesystem twice on same mountpoint (3394)
5031 - allow llmount to create routes for mounting behind routers (3320)
5032 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5033 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5034 - fix rare problem with rename on htree directories (3417)
5036 - allow default OST striping configuration per directory (1414)
5037 - fix compilation for qswnal for 2.6 kernels (3125)
5038 - increase maximum number of MDS request buffers for large systems
5039 - change liblustreapi to be useful for external progs like lfsck (3098)
5040 - increase local configuration timeout for slow disks (3353)
5041 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5043 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5046 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5047 - don't overwrite extent policy data in reply if lock was blocked (2901)
5048 - drop filter export grants atomically with removal from device (2663)
5049 - del obd_self_export from work_list in class_disconnect_exports (2908)
5050 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5051 - swab reply message in mdc_close, other PPC fixes (2464)
5052 - fix destroying of named logs (2325)
5053 - overwrite old logs when running lconf --write_conf (2264)
5054 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5055 - fix race in target_handle_connect (2898)
5056 - mds_reint_create() should take same inode create lock (2926)
5057 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5058 - don't close files for self_export to avoid uninitialized obd (2936)
5059 - allow MDS with the same name as client node (2939)
5060 - hold dentry reference for closed log files for unlink (2325)
5061 - reserve space for all logs during transactions (2059)
5062 - don't evict page beyond end of stripe extent (2925)
5063 - don't oops on a deleted current working directory (2399)
5064 - handle hard links to targets without a parent properly (2517)
5065 - don't dereference NULL lock when racing during eviction (2867)
5066 - don't grow lock extents when lots of conflicting locks (2919)
5068 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5071 - account for cache space usage on clients to avoid data loss (974)
5072 - lfsck support in lustre kernel code (2349)
5073 - reduce journal credits needed for BRW writes (2370)
5074 - orphan handling to avoid losing space on client/server crashes
5075 - ptlrpcd can be blocked, stopping ALL progress (2477)
5076 - use lock value blocks to assist in proper KMS, faster stat (1021)
5077 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5078 - recovery for initial connections (2355)
5079 - fixes for mds_cleanup_orphans (1934)
5080 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5081 - block all file creations until orphan recovery completes (1901)
5082 - client remove rq_connection from request struct (2423)
5083 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5084 - recovery timer race (2670)
5085 - mdc_close recovey bug (2532)
5086 - ptlrpc cleanup bug (2710)
5087 - mds timeout on local locks (2588)
5088 - namespace lock held during RPCs (2431)
5089 - handle interrupted sync write properly (2503)
5090 - don't try to handle a message that hasn't been replied to (2699)
5091 - client assert failure during cleanup after abort recovery (2701)
5092 - leak mdc device after failed mount (2712)
5093 - ptlrpc_check_set allows timedout requests to complete (2714)
5094 - wait for inflight reqs when ptlrpcd finishes (2710)
5095 - make sure unregistered services are removed from the srv_list
5096 - reset bulk XID's when resending them (caught by 1138 test)
5097 - unregister_bulk after timeout
5098 - fix lconf error (2694)
5099 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5100 - readahead locks pages, leaves pending causing memory pressure (2673)
5101 - increase OST request buffers to 4096 on large machines (2729)
5102 - fix up permission of existing directories in simple_mkdir (2661)
5103 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5104 - don't assign transno to errored transactions (2742)
5105 - don't delete objects on OST if given a bogus objid from MDS (2751)
5106 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5107 - if rq_no_resend, then timeout request after recovery (2432)
5108 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5109 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5110 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5111 - MDS recovery shouldn't delete objects that it has given out (2730)
5112 - if enqueue arrives after completion, don't clobber LVB (2819)
5113 - don't unlock pages twice when trigger_group_io returns error (2814)
5114 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5115 - don't write pages to disk if there was an error (1450)
5116 - don't ping imports that have recovery disabled (2676)
5117 - take buffered bytes into account when balancing socknal conn (2817)
5118 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5119 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5120 - remove little-endian swabbing of llog records (1987)
5121 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5122 - retry reposting request buffers if they fail (1191)
5123 - grow extent at grant time to avoid granting a revoked lock (2809)
5124 - lock revoke doesn't evict page if covered by a second lock (2765)
5125 - disable VM readahead to avoid reading outside lock extents (2805)
5127 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5128 - updated kernel patches for hp-2.4.20 kernel (2681)
5130 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5133 - fix truncated write corruption (2366)
5134 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5136 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5137 - ensure len(uuid) < 37 in lmc (1171)
5138 - fix ia64 OOPS in llog_test (2255)
5139 - zero end of page at obdfilter for partial page writes (2648)
5140 - don't leave stale dentries around after renames (bug 2428)
5141 - fix timeouts when evicting a client with a single lock held (2642)
5142 - set deadline for the initial HELLO message to drain (2634)
5143 - print out dotted-quad IP addresses in the socknal (2302)
5145 - additional debugging for MDS client eviction problem (2443)
5146 - fix mkfsoptions support for osts (2603, 2604)
5148 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5151 - add series for the vanilla 2.6.0 kernel
5152 - add series for the vanilla 2.4.24 kernel
5153 - add series for a cray x86/64 UL kernel drop
5154 - fix xattr patches for the vanilla 2.4.19 series
5156 - generate true UUIDs in lmc (1171)
5157 - have portals stack dumping break in UML (2466)
5158 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5159 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5160 - wake callback waiting threads on client eviction (2460)
5161 - Add --ptldebug and --subsystem to lmc (1719)
5162 - update assertion to allow safe interrupt allocation
5163 - set rq_no_resend for cancel requests (2432)
5164 - recalculate ptlrpcd timeout after resend (2494)
5165 - call vfs_rmdir when removing pending directories (2368)
5166 - fix renaming a file to itself (2429)
5167 - lmc creates a default one-stripe lov (2454)
5168 - expand procfs space to handle large clusters (2326)
5169 - increase UML stack to avoid overflow
5170 - update lconf's list of debug and subsystem masks
5171 - fix lfs find --obd (2510)
5172 - /proc tunable for disabling filter read caching (2591)
5173 - stop rpm packages from altering slapd.conf (2301)
5174 - disable nagle in the socknal under 0conf (2578)
5175 - choose mds inode size based on stripe count (2572)
5176 - fix kernel-source rpm problems (2516)
5178 - add --disable-doc to avoid pdf generation (2421)
5179 - update documentation, tests, type-os, comments
5180 - avoid format warnings on ia64
5181 - remove the TOE NAL
5182 - tiny code cleanups by removing unused fields
5184 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5187 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5188 - remove the most blatant lies from BUILDING file (2371)
5189 - change default debug level to reasonable production setting
5190 - reduce client side cache size to reduce cache flush time
5191 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5192 - make TCP zerocopy and pinger support enabled by default (2476)
5193 - sync writes completed after process exits caused crashes (2319)
5194 - maintain correct mount count on the MDS (2356)
5195 - backout 1557, because 2316 wasn't really fixed
5196 - better file I/O statistics gathering in /proc
5197 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5198 - another kernel patch to fix zero-copy TCP function export
5199 - don't take duplicate lock when processing re-sent getattr (2420)
5200 - lctl uses obd_self_export instead of creating new conn (2353)
5201 - MDS/OST recovery case which requires object creation asserted (2425)
5202 - move lfs from /usr/sbin to /usr/bin in packages
5203 - fix race between mds_client_add and mds_client_free (2417)
5204 - use kmalloc instead of slabs in portals (2430)
5205 - don't create duplicate records when a failover MDS is present (2442)
5206 - remove unnecessary mount age check (2332)
5207 - don't remove directory inodes from locks prematurely (2451)
5208 - don't break if MDS service name is the same as hostname (2103)
5209 - fix races in client write RPC generation when cache full (2482)
5211 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5214 - remove now-unused request->rq_obd (278)
5215 - if an allocation fails, print out how much memory we've used (1933)
5216 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5217 - add the "configurable stack size" patch to most series files (1256)
5218 - ability to write large log records, for 100+ OST configs (2306)
5219 - fix NULL deref when filter_prep fails (2314)
5220 - fix operator precedence error in filter_sync
5221 - dynamic allocation of socknal TX descriptors (2315)
5222 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5223 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5224 - ensure that CWARN messages in Portals always get to the syslog
5225 - __init/__exit are not for prototype decls (ldlm_init/exit)
5226 - x86-64 compile warning fixes
5227 - fix gateway LMC keyword conflict (2318)
5228 - fix MDS lock inversions in getattr/reint paths (1844)
5229 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5230 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5231 - DLM race condition prevented some lock evictions (2328)
5232 - ENOMEM detection and retry on socknal sends (2230)
5233 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5234 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5235 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5236 - better error messages when a client is rejected during recovery (1505)
5237 - avoid cancelling locks which were never granted, after failure (2330)
5238 - fix i_sem/journal inversion in mds_client_add (2333)
5239 - fix truncate/getattr lock cycle deadlock (2334)
5240 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5241 - fix two rare exit paths which could leak an l_lock() ref (2321)
5242 - fencepost error in MDS/OST orphan recovery (2226)
5243 - make log record alignment 8 bytes (1988)
5244 - lstripe now fails when requested offset > ost_count (2237)
5245 - ensure that all kernel series have a complete list.h (1607)
5246 - fix crashes in special-file operations (2316)
5247 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5249 - allow configurable automake binary, for testing new versions
5250 - small update to the lfs documentation
5252 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5254 * fix negative export reference count in fsfilt_sync (2312)
5256 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5257 * release candidate 0.9.1
5259 - orphans are moved into the PENDING directory for possible recovery
5260 - replayed opens now open by fid for orphan/rename safety (1042)
5261 - last close of an orphan inode generates a transno (683)
5262 - chdir() and mount() now pin the directory entry (1020)
5263 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5264 - discard very old requests without processing them (1502)
5265 - remove some common, well-understood CERRORs (1505)
5266 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5267 - clear "grant" flags in OST replies until OST grant code lands (1644)
5268 - fix read performance by not clobbering i_blksize on client (1598)
5269 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5270 - make LRU size a /proc tunable, clears locks when reduced (707)
5271 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5272 - server locks take references on exports now (1558)
5273 - build fixes for 2.4.20-rh trees (1663)
5274 - return an error from lov_create if all OSCs are inactive (1751)
5275 - fix import levels when a reconnect happens without a timeout (1597)
5276 - exit early from mds_open if we get a lookup error (1749)
5277 - partial page read at EOF wouldn't wait for disk before sending (1642)
5278 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5279 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5280 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5281 - write/truncate lock inversion (1639)
5282 - Don't auto-load obdclass, portals modules during cleanup (1495)
5283 - fix timestamps from jumping to "now" (1763)
5284 - extra journal assertions (1648)
5285 - add an extra multiunlink test (1771)
5286 - fix read_record/write_record API (1776)
5287 - fix leak of offset_extent, possible incorrect i_size later (1772)
5288 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5289 - Debugging for the kqswnal_get_idle_tx problems (1820)
5290 - Allow recovery to be attempted multiple times (1536)
5291 - Write out MDS last_rcvd file after it is first created (1600)
5292 - Fix tx_descriptor leak in failed transmit situations (1827)
5293 - ext3 journaling fixes for assertion failure after IO error (1871)
5294 - class_export_put() on freed export after completion AST error (1896)
5295 - Fix revalidate looping in VFS (1322)
5296 - Don't access a freed export during MDS_REINT timeout (1521)
5297 - Add open-unlink recovery support on the MDS (1673,1764)
5298 - Return an error if no MDS data was read from last_rcvd (1946)
5299 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5300 - Don't setup a disk device that doesn't match exported UUID (317)
5301 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5302 - avoid committing NULL handle in force close
5303 - local.sh is now a one-stripe LOV configuration
5304 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5305 - fix ext3 htree duplicate directory entry corruption (1516)
5306 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5307 - update to new LOV EA format (2097)
5308 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5310 2003-06-15 Phil Schwan <phil@clusterfs.com>
5313 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5314 - per-import/export recovery handling (958, 931, 959)
5315 - multiple last-rcvd slots, for serving multiple FSes (949)
5316 - connections are again shared between multiple imp/exports (963, 964)
5317 - "umount -f" would hang if any requests needed to be sent (393, 978)
5318 - avoid pinning large req buffer by copying for queued messages (989)
5319 - add "uuid" to "lctl device" command to help upcalls (991)
5320 - "open" RPCs with transnos would confuse recovery counters (1037)
5321 - do proper endian conversion of all wire messages (288, 340, 891)
5322 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5323 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5324 - fix ptlrpc_connection leak in target_handle_connect (1174)
5325 - fix import refcounting bug in OST and MDS cleanup (1134)
5326 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5327 - fix very unlikely obd_types race condition (501)
5328 - remove osc_open hack for echo_client (1187)
5329 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5330 - a failure in read_inode2 leads to deadlock (1139)
5331 - cancel ack-locks as soon as transaction is committed (1072)
5332 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5333 - make sure to commitrw after any preprw to avoid deadlock (1162)
5334 - failing to execute a file in a lustre FS would lock inode (1203)
5335 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5336 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5337 - obd_uuid2tgt would walk off the end of the list (1255)
5338 - on IA64 the osc would give portals incorrect bulk size (1258)
5339 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5340 - fix lock inversion caused by new llite matching code (1282)
5341 - limit the number of dirty pages on a client to 10MB (1286)
5342 - timed out locks were not being corrected cancelled (1289)
5343 - fix O_DIRECT above 4GB on IA-32 (1292)
5344 * major user-visible changes
5345 - fail out/fail over policy now controlled by the upcall (993)
5347 - add OBD_PING to check server availability and failure (954)
5348 - lustre messages are now sent in sending host order (288, 340, 891)
5349 - add eadatalen to MDS getattr reply (340)
5350 - OST read replies may contain second buffer, with per-page status (593)
5352 2003-03-11 Phil Schwan <phil@clusterfs.com>
5355 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5356 - failed open()s could cause deadlock; fixed (867, 869)
5357 - stop cancelling OST locks when files are closed (481)
5358 - overlapping XID spaces caused network corruption (851, 853)
5359 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5360 - setattr_raw added, to do single-RPC, server-side setattrs
5361 - lmc/lconf syntax change for OST UUIDs
5362 - fix crashy race condition between ptlrpc_free_req and osc_close
5363 - don't use request in mdc_enqueue if we hit a timeout (889)
5364 - don't set the inode i_size for regular files from the MDS (896)
5365 - handle out of order completion AST (842)
5366 - don't LBUG if a lock request times out after receiving AST (913)
5367 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5368 - if a bad lock AST arrives, send an error instead of dropping entirely
5369 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5370 - fix leak in bulk IO when only partially completed (899, 900, 926)
5371 - fix O_DIRECT for ia64 (55)
5372 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5373 - C-z after timeout could hang a process forever; fixed (977)
5375 - client-side I/O cache (678, 924, 929, 941, 970)
5377 - READPAGE and SETATTRs which don't take server-side locks get
5380 2003-02-11 Phil Schwan <phil@clusterfs.com>
5383 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5384 - Fix fsfilt_extN_readpage() to read a full page of directory
5385 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5386 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5387 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5388 - Most intent operations were removed, in favour of a new RPC mode
5389 that does a single RPC to the server and bypasses most of the VFS
5390 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5391 - Aggressively cancel local locks on DLM servers
5392 - mds_reint_unlink sends EA to the client if it's the last nlink.
5393 client uses that EA to unlink OST objects.
5394 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5395 - recursive symlinks were fixed (439)
5396 - fixed NULL deref in DEBUG_REQ
5397 - filter_update_lastobjid no longer calls sync, which annoyed extN
5398 - fixed multi-client small-writes to a single file problem (445)
5399 - fixed mtime updates during file writes (607)
5400 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5401 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5402 - fixed Linux OST size reporting problem (444, 656)
5403 - OST now updates object mtime with writes or setattr (607, 619)
5404 - client verifies file size before zeroing page past EOF (445)
5405 - OST now writes last allocated objid to disk with allocation (108)
5406 - LOV on echo now works (409)
5408 - mds_reint_unlink sends a new buffer, with the EA included. this
5409 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5410 set if a regular file was being unlinked, and it was the last link
5411 - use PtlGet from the target for bulk writes (315)
5412 - OST now updates object mtime with writes or setattr (607, 619)
5413 - LDLM now has a grant-time callback to revalidate locked items, if
5415 - Many MDS operations were reorganized to combat race conditions
5417 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5418 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5420 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5423 - Fully reactivate OST imports after reconnection (512, others)
5424 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5425 - More graceful error handling for truncating on dead OST (515)
5426 - Don't error out unless we're actually accessing dead stripes (474)
5427 - Fix garbage sizes when stripes are missing (410)
5428 - LRU counters were broken, causing constant lock purge (433, 432)
5429 - garbage on read from stripes with failed OSTs (441)
5430 - mark OSCs as active before reconnecting during recovery (438)
5431 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5432 - lfind did not preserve OST order in output (443)
5433 - symlinks cause hung clients, incorrect data (439)
5434 - stop dereferencing request after dropping refcount (457)
5435 - don't LASSERT(spin_is_locked) on non-SMP (455)
5436 - fixes for many rename() bugs
5437 - fstat didn't correctly synchronize attributes (399)
5438 - server must handle lock cancellation during blocking AST prep (487)
5439 - bulk descriptors were free()d too soon (511)
5440 - fix paths in lconf, which would load incorrect modules (451, 507)
5441 - fix confusing lconf 'host not found' error message (386)
5442 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5443 - fix race condition in mdc_blocking_ast() for inode access (526)
5444 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5445 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5446 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5447 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5448 - fix lstripe in several ways (kernel side) (527)
5449 - fix request leak in ldlm_cli_enqueue (262)
5450 - incorrect OSC was marked inactive after OST failure
5451 - call mds_fs_cleanup before unmounting filesystem (524)
5452 - fix races between taking ns_lock and ldlm_lock_change_resource
5453 - fix races updating LOV export open file list
5454 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5455 - fix recovery NULL deref in ldlm_cli_cancel_unused
5456 - fix some DLM races by using new hash table for lock handles (419)
5457 - permit the client to specify desired inodes, at replay
5458 - duplicate requests when we queue them for replay reintegration
5459 - fix last_rcvd offset calculation
5460 - sync after each recovered transaction, so we always make progress
5461 - never, not always, ERESTART requests without transnos
5462 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5464 - skip replay if the MDS says that the client is already connected
5465 - don't check for a recovery-enabled export to match lctl's UUID
5466 - don't INC_USE_COUNT for phantom exports
5467 - don't crash when cleaning up phantom exports (567)
5468 - don't double-finish or set replay data for errored mdc_open requests
5469 - abort requests when they time out, so we don't get old replies
5470 - send/receive replies for AST messages again
5471 - if the client says that it doesn't have the lock, cancel it on the
5473 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5474 mark it as destroyed, it will all work out when decref is called
5475 - fix module use counts (22, 581)
5477 - ASTs now expect a reply (server cancels lock on error reply)
5479 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5482 - fix many simultaneous client startup (392)
5483 - fix dentry->d_it clobbering
5484 - credentials weren't being shipped for readdir/getattr operations
5485 - remove invalid assertions triggered during some concurrent MD
5487 - proper Lustre versions added (336, 389)
5488 - fix memory leak for create error case (398)
5489 - fix LOV locking bug that would get cli/srv out of sync
5490 - fix echo client over LOV (409)
5491 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5492 - fix double-O_EXCL intent crash (424)
5493 - avoid sending multiple lock CANCELs (352)
5495 - MDS can do multi-client recovery (modulo bugs in new code)
5497 - many updates, edits, cleanups
5499 2002-11-18 Phil Schwan <phil@clusterfs.com>
5502 - fix null d_it dereference (346)
5503 - fix full OST/dbench hang (333)
5504 - fix permission problem with file removal (286)
5505 - fix removal of OSCs from LOV when they fail
5506 - fix NULL deref during bulk timeout (214)
5507 - fix problems related to multiple filesystems on one MDS (241)
5508 - fixed serious subtle metadata locking bugs
5509 - free locks on clients when inodes are removed due to memory
5511 - fix inode pointer in lock data (285)
5512 - partial support for multiple MDS on a single host (241)
5513 - data locks weren't cancelled at clear_inode time (290, 311)
5514 - intent locks could lead to unbounded lock growth (205)
5515 - added a maximum lock count, an LRU list, and a flusher
5516 - fix multiple rename (365)
5517 - properly abstracted the echo client
5518 - OSC locked 1 byte too many; fixed
5519 - rewrote brw callback code:
5520 - fixed recovery bugs related to LOVs (306)
5521 - fixed too-many-pages-in-one-write crash (191)
5522 - fixed (again) crash in sync_io_timeout (214)
5523 - probably fixed callback-related race (385)
5525 - Add capability to MDS protocol
5526 - LDLM cancellations and callbacks on different portals
5528 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5531 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5532 - defer open object destruction to close time (601981, 138)
5533 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5534 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5535 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5536 - don't call obd_setattr in ll_file_release for destroyed objects
5537 * protocol change to lustre_msg: move |version| and add |flags|
5538 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5539 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5540 * added replay of create, unlink, link and rename operations during
5541 MDS failover; recovery should be much more robust now
5542 * remove failed OSCs from LOVs (only lov_create uses this so far)
5543 * the lustre-HOWTO was brought (more) up to date (582544)
5545 2002-10-23 Phil Schwan <phil@clusterfs.com>
5548 - in-use dentries weren't being reused properly (617851)
5549 - prevent multiple LDLM setup (599178)
5550 - fix LOV size calculations for truncate (617853)
5551 - fix client handling of MDS intent errors (POSIX)
5552 - fix permission bug in lovstripe.c test (624321)
5553 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5554 - truncate past end of file could corrupt data
5555 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5556 - a race in recovery could return ETIMEDOUT to apps (623947)
5557 - building outside the source directory was fixed
5558 * the lustre-HOWTO was brought (more) up to date (582544)
5559 * major progress was made on recovery functionality
5561 2002-10-10 Phil Schwan <phil@clusterfs.com>
5564 - recovery deadlock fix
5565 - rm -rf causes LBUG fix (617817)
5566 - file open by multiple tasks fix (618962)
5567 - directory permissions bugs (602707 and 620007)
5568 - journal_stop fixed with locking (611313)
5569 - O_APPEND failures resolved (618273, perhaps 614459)
5570 - lconf PATH fix (619770)
5571 - IA64 build fix (621450)
5572 - RPC buffer sizes scale with amount of memory
5574 2002-10-01 Phil Schwan <phil@clusterfs.com>
5577 - locks would be cancelled without throwing away data pages,
5578 resulting in inconsistent data (605627)
5579 - inode attributes were not always being refreshed (605627, 612449)
5580 - lconf now continues to cleanup after lctl reports an error
5581 - MDS now enforces user permissions (602707)
5582 - lprocfs cleanup fixed, but not yet enabled (614157)
5583 - fixed infinite server hang, should a client not respond to an AST
5584 - avoid going into recovery if user calls readlink() with a buffer
5585 that's too small (613941)
5586 - AST RPCs no longer require replies (614867) -- this may be changed
5587 - don't crash server if client sends an IOV that's too big (611336)
5588 - fixed lock conversion deadlock (611892)
5589 - fixed the following of symlinks (614622)
5590 * recovery: the server can remove locks from a client that dies, other
5591 clients can make progress
5592 * more extN patch fixes
5593 * compile-time configurable ptlrpc buffer allocations
5595 - collaborative read cache document
5596 - Lustre Lite Performance CDR document-in-progress
5598 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5601 - fix typo in patch-2.4.18
5603 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5606 - clear ptlrpc request each time in handle_incoming_request()
5607 - unlink of files now destroys the object on the OST
5609 2002-09-19 Peter Braam <braam@clusterfs.com>
5611 * add hard link support
5612 * change obdfile creation method
5613 * kernel patch changed
5615 2002-09-19 Peter Braam <braam@clusterfs.com>
5618 - stack overflow bug in extN fixed
5620 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5622 * documentation updates
5623 - add man pages for config tools
5624 - update tests/README to describe testing with new config tools
5625 - finish metadata API descriptions
5626 * bug fixes and cleanups
5627 - statfs workaround for 16TB limit
5628 - LOV stripe allocation improved, can stripe on subset of OSTs
5629 - LOV file size/IO offset was wrong for files > 4GB in size
5630 - object EA data was being dropped, caused files to be unreadable
5631 - memory overflow with non-LOV OST caused memory corruption
5632 - fixed regression tests to work with new config tools, obdfilter
5633 - fixed bug when directory size became larger than 1 block
5634 - fixed bug (for single client case) when PWD was deleted
5635 - invalidate local directory pages when doing intent-based ops
5636 - avoid LDLM oops when lock callback contained bad data
5638 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5640 * documentation updates
5641 * bug fixes and cleanups
5642 - configuration tools
5645 - 64-bit compile warnings
5646 - 64-bit internal statfs data
5648 * test_brw on persistent OST devices
5650 * lprocfs (disabled)
5652 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5654 * documentation updates
5655 * bug fixes and cleanups
5656 * configuration tools
5658 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5661 * many small fixes to 0_5_4
5662 * io/network handling
5663 * thinkos in MDS operations
5665 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5668 * crucial basic fixes to 0.5.3
5669 * IOR, Iozone work over Elan
5672 2002-08-07 Phil Schwan <phil@clusterfs.com>
5673 * version 0_5_3, our first alpha
5674 * we use the new Portals iovs
5675 * documentation updates
5676 * bug fixes and cleanups
5677 * small changes in the DLM wire protocol
5679 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5680 * version 0_5_1 with some initial stability,
5681 * locking on MD and file I/O.
5682 * documentation updates
5683 * several bug fixes since 0.5.0
5684 * small changes in wire protocol
5686 2002-07-18 Phil Schwan <phil@clusterfs.com>
5688 * delivered as Lustre Light Alpha
5689 * fixed a crash after handling invalid MDS requests
5690 * fixed directory pages for architectures with non-4k pages sizes
5692 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5693 * release version v0_4_4
5694 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5696 * Fixes a number of interruption problems with OST operations.
5697 * Update documentation for portals header changes
5698 * Move all wire protocol structs/defines to lustre_idl.h
5699 * Fixes symlink length bug.
5700 * Add tcpdump to repository.
5702 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5703 * release version v0_4_3
5704 * Fixes statfs for inodes on extN.
5705 * Fixes bug in runtests which would delete /etc/hosts.
5706 * Use 64-bit object IDs wherever possible (not into VFS though)
5707 Remove ost_get_info, which is unused by lustre, and out of date.
5709 2002-07-03 Peter Braam <braam@clusterfs.com>
5710 * release version v0_4_2 Fixes a lookup error (type not passed)
5711 * move forward to head of Portals
5712 * move forward to latest Lustre kernel
5714 2002-06-25 Peter Braam <braam@clusterfs.com>
5715 * release version v0_4_1. Hopefully stable on single node use.